109 - UnivalentfunctionsandTeichmuellerSpaces PDF

Olli Lehto
Univalent Functions
and Teichmller Spaces
With 16 Illustrations
Springer- Verlag
Norid Publishing Corp
0(11 Lehto
Departme'nt of Mathematics
University of Helsinki
00100 Helsinki
Finland
Editorial Board
P. R. Halmos
Department of Mathematics
Santa Clara University
Santa Clara, CA 95053
F. W. Gehring
Department of Mathematics
University of Michigan
Ann Arbor, MI 48109
U.S.A.
U.S.A.
AMS Subject Classirications: 30-01, 32615, 30F30. 30F35
Library of Congress Cataloging in Publication Data

Lehto, 0111.
Univalent functions and Teichynller spaces.
(Gi'aduate texts in
109)
Bibliography: p.
Includes index.
I. Univalent functions. 2. Teichmller spaces.
3. Riemann surfaces. I. Title. II, Series.
QA331.L427
1986
515
86-4008
1987 by Springer-Verlag New York
Inc.
All rights reserved. No part of this book may be translated or reproduced in any form without
writtcn permission from Springer-Verlag, 175 Fifth Avenue, New York. New York 10010, U.S.A.
Reprinted by World Publishing Corporation, Beijing, 1990

for distribution and sale in The People's Republic of China only
ISBN 750620732--X
ISBN 0-387-96310-3 Springer-VerLag New York Berlin Heidelberg
ISBN 3-540-96310-3 Springer-Verlag I3cilin Heidelberg New York
Preface
This monograph grew out of the notes relating to the lecture courses that I
gave at the University of Helsinki from 1977 to 1979, at the Eidgenossische
Technische Hochschule Znch in 1980, and the University of Minnesota
in 1982. The book presumably would never have been written without Fred
Gehrfng's continuous encouragement. Thanks to the arrangements made by
Edgar Reich and David Storvick, I was able to spend the fall term of 1982 in
Minneapolis and do a good part of the writing there. Back in Finland, other
commitments delayed the completion of the text.
At the final stages of preparing the manuscript, I was assisted first by Mika
Seppla and then by Jouni Luukkainen, who both had a grant from the
Academy of Finland. I am greatly indebted to them for the improvements
they made in the text.
1 also received valuable advice and criticism from Kari Astala, Richard
Fehlmann, Barbara Flinn, Fred Gchring, Pentti Jrvi, Irwin Kra, Matti
Lehtinen, lippo Louhivaara, Bruce Palka, Kurt Strebel, Kalevi Suominen,
Pekka Tukia and Kalle Virtanen. To all of them I would like to express my
gratitude. Raili Pauninsalo deserves special thanks for her patience and great
care in typing the manuscript.
Finally, I thank the editors for accepting my text in Springer-Verlag's wellknown series.
Helsinki, Finland
June 1986
Oh Lehto
Contents
Preface
Introduction
v
.
CHAFFER I
Quasiconformal Mappings
Introduction to Chapter I
1. Conformal Invariants
1.1 Hyperbolic metric
1.2 Module of a quadrilateral
1.3 Length-area method.
1.4 Rengel's inequality
1.5 Module of a ring domain
1.6 Module of a path family
2. Geometric Definition of Quasiconformal Mappings
2.1 Definitions of quasloonformality
2.2 Normal families of quasiconformal mappings
2.3 Compactness of quasiconformal mappings
2.4 A distortion function
2.5 Circular distortion
3. Analytic Definition of Quasiconformal Mappings
3.1 Dilatation quotient
3.2 Quasiconlurwal diffetorphisms
3.3 Absolute noutinuity and differentiability
3.4 Generalized derivatives
3.5 Analytic characterization of quasiconformality
4. Seltrami Differential Equation
4.1
dilatation
4
5
5
7
8
9
10
11
12
12
13
14
is
17,
18
18
19
20
21
22
23
23
viii
Contents
0
4.2
Quasiconformal mappings and the Beltrami equatior
24
25
26
4.3 Singular integrals
4.4 Representation of quasiconformal mappings

4.5 Existence theorem
4.6 Convergence of complex dilatations
4.7 Decomposition of quasiconformal mappings
5. The Boundary Value Problem
5.1 Boundary function of a quasiconformal mapping
5.2 Quasisymmetric functions
5.3 Solution of the boundary value problem
5.4 Composition of BeurlingAhlfors extensions
5.5 Quasi-isometry
5.6 Smoothness of solutions
5.7 Extremal solutions
6. Quasidiscs
6.1 Quasicircles
6.2 Quasiconformal reflections
6.3 Uniform domains
6.4 Linear local connectivity
6.5 Arc condition
6.6 Conjugate quadrilaterals
6.7 Characterizations of quasidiscs
27
28
29
30
30
31
33
34
35
36
37
38
38
39
41
44
45
46
47
CHAPTER II
Univalent Functions
50
Introduction to Chapter II
I. Schwarzian Derivative
1.1 Definition and transformation rules
1.2 Existence and uniqueness
1.3 Norm of the Schwarzian derivative
1.4 Convergence of Schwarzian derivatives
1.5 Area theorem
1.6 Conformal mappings of a disc
2. Distance between Simply Connected Domains
2.1 Distance from a disc
2.2 Distance function and coefficient problems
2.3 Boundary rotation
2.4 Domains of bounded boundary rotation
2.5 Upper estimate for the Schwarzian derivative
2.6 Outer radius of univalence
2.7 Distance between arbitrary domains
3. Conformal Mappings with Quasiconformal Extensions
3.1 Deviation from Mbius transformations
3.2 Dependence of a mapping on its complex dilatation
3.3 Schwarzian derivatives and complex dilatations
3.4 Asymptotic estimates. .
3.5 Majorant principle
3.6 Coefficient estimates
50
51
51
53
54
55
58
59
60
60
61
62
63
65
66
67
68
68
69
72
73
76
77
Contents
ix
4
(Jnivalence and Quasiconforrnal Extensibility of Meromorphic

Functions
4.1 Quasiconformal reflections under Mbius transformations
4.2 Quasiconformal extension of conformal mappings
4.3 Exhaustion by quasidiscs
4.4 Definition of Schwarzian domains
4.5 Domains not linearly locally connected
4.6 Schwarzian domains and quasidiscs
5. Functions Univalent in a Disc
5.1 Quasiconformal extension to the complement of a disc
5.2 Real analytic solutions of the boundary value problem
5.3 Criterion for univalence
5.4 Parallel strips
5.5 Continuous extension
5.6 Image of discs
5.7 Homeomorphic
4.
79
79
81
83
83
84
86
87
87
89
89
90
91
92
94
CHAPTER III
Universal Teichmiiller Space

Introduction to Chapter III
1. Models of the Universal Tcichmller Space
I.! Equivziknt
mappings
1.2 Group structures
1.3 Normalized conformal mappings
1.4 Sewing problem
1.5 Normalized quasidiscs
2. Metric of the Universal Teichmlier Space
2.1 Definition of the Teichmller distance
2.2 Teichmller distance and complex dilatation
2.3 Geodesics for the Teichmller metric
2.4 Completeness of the universal TeichmUller space
3. Space of Quasisymmetric Functions
3.1 Distance between quasisymxnetric functions
3.2 Existence of a section
3.3 Contractibility of the universal
space
3.4 Incompatibility of the group structure with the metric
4. Space of Schwarzian Derivatives
4.1 Mapping into the space of Schwarzian derivatives
4.2 Comparison of distances
4.3 Imbedding of the universal Teichmller space
4.4 Schwarzian derivatives of univalent functions
4.5 Univalent functions and the universal Teichmiiller space
4.6 Closure of the universal Teichmller space
5. Inner Radius of Univalence
5.1 Definition of the inner radius of univalence
5.2 Isomorphic TekhmUller spaces
5.3 Inner radius and quasiconformal extensions
96
96
97
97
98
99
100
101
103
103
105
106
108
108
109
110
111
111
112
113
115
116
116
118
118
119
120
Contents
5.4 Inner radius and quasiconformal reflections

5.5 Inner radius of sectors
5.6 Inner radius of ellipses and polygons
5.7 General estimates for the inner radius
121
122
125
126
CHAPTER IV
Ricmann Surfaces
Introduction to Chapter IV
128
128
129
129
130
1. Manifolds and Their Structures

Li Real manifolds
1.2 Complex analytic manifolds
1.3 Borderofasurface
1.4 Differentials on Ricmann surfaces
1.5 Isothermal coordinates
1.6 Riemann surfaces and quasiconformal mappings
2. Topology of Covering Surfaces
2.1 Lifting of paths
2.2 Covering surfaces and the fundamental group
2.3 Braflched covering surfaces
2.4 Covering groups
2.5 Properly discontinuous groups
3. Uniformization of Riemann Surfaces
3.1 Lifted and projected conformal structures
3.2 Riemaun mapping theorem
3.3 Representation of Riemann surfaces
3.4 Lifting of continuous mappings
3.5 Homotopic mappings
3.6 Living of differentials
4. Group. of Mbius Transformations
4.1 Covering groups acting on the plane
4.2
groups
4.3 Elementary groups
4.4 ICisiniangroups
4.5 Structure of the limit set
4.6 InvarIant domains
5. Compact Riemana Surfaces
5.1 Covering groups over compact surfaces
5.2
of a compact surface
5.3 FunctIon theory on compact Riemann surfaces
siufaces
5.4 DivIsors on
theorem
5.5
of Quadratic Differentials
6.1 Natural parameters
6.2 Straigbt line, and trajectories
63
Oiimm*ation
,(
6.4 Trajeciodes in the large
131
132
133
134
135
135
136
137
.
138
140
142
142
143
144
145
146
147
149
149
150
151
152
153
155
157
157
-...
.
...
158
158
t59
160
161
161
163
164
165
Contents
6.5 Periodic trajectories

6.6 Non-periodic trajectories
7. Geodesics of Quadratic Differentials
7.! Definition of the induced metric
7.2 Locally shortest curves
7.3 Geodesic polygons
7.4 Minimum property of geodesics
7.5 Existence of geodesics
7.6 Deformation of horizontal arcs
Xi
166
166
168
168
169
170
171
173
174
CHAPTER V
TeichmUller Spaces
Introduction to Chapter V
1. Quasiconformal Mappings of Riemann Surfaces
1.1 Complex dilatation on Riemann surfaces
1.2 Conformal structures
1.3 Group isomorphisms induced by quasiconformal mappings
1.4 Homotopy modulo the boundary
1.5 Quasiconformal mappings in homotopy classes
2. Definitions of Teichmller Space
2.1 Riemann space and Teichmller space
2.2 Teichmller metric
2.3 Teichmller space and Beltrami differentials
2.4 Teicbmller space and conformal structures
2.5 Conformal structures on a compact surface
2.6 Isomorphisms of Teichmuller spaces
2.7 Modular group
3. Teichmller Space and Lifted Mappings
3.1 Equivalent Beltrami differentials
3.2 Teichmflller space as a subset of the universal space
3.3 Completeness of Teichmller spaces
3.4 Quasi-Fucbsian groups
3.5 Quasiconformal reflections compatible with a group
3.6 Quasisymmetric functions compatible with a group
3.7 Unique extremality and Teichmller metrics
4. Teichmller Space and Schwarzian Derivatives
4.1 Schwarzian derivatives and quadratic differentials
4.2 Spaces of quadratic differentials
4.3 Schwarzian derivatives of univalent functions
4.4 Connection between Teichmfillcr spaces and the universal space....
4.5 Distance to the boundary
4.6 Equivalence of metrics
4.7 Ben unbedding

4.8 Quasiconformal extensions compatible with a group
5. Complex Structures on Teichmller Spaces
5.1 Holomorphic functions in Banach spaces
175
175
176
176
178
178
180
181
182
182
183
184
185
186
187
188
189
189
190
190
191
192
193
195
196
196
197
197
198
200
201
203
205
205
Contents
xii
5.2 Banach manifolds

5.3 A
mapping between Banach spaces
5.4 An atlas on the Teichmtiller space
5.5 Complex analytic structure
5.6 Complex structure under quasiconformal mappings
6. Teichmiiller Space of a Torus
6.1 Covering group of a torus
6.2 Generation of group isomorphisms
6.3 Conformal equivalence of tori
6.4 Extremal mappings of tori
6.5 Distance of group isomorphisms from the identity
6.6 Representation of the Teichmller space of a
6.7 Complex structure of the Tek.lmiller space of torus
7. Extremal Mappings of Riemann Surfaces
7.1 Dual Banach spaces
7.2 Space of integrable holoniorphic quadratic differentials
7.3 Poincar theta series
7.4 !nfinitesimaHy trivial difIerentials
7.5 Mappings with infinitesimally trivial dilatat ions
7.6 Complex dilatations of extremal mappings
7.7 Ieichmllcr
7.8 Extrernal mappings of compact surfaces
8. Uniqueness of Extremal Mappings of Compac.t
8.1 TeichmUlier mappings and quadratic diflerenijais
8.2 Local representation of Teichrniil!cr mappings
8.3 Stretching function and the Jacobian
8.4 Avcmge stretching
8.5 Teichniller's uniqueness theorem
9. Teichmiiller Spaces of Compact Surfaces
9.1 Teichmller imbcdding
9.2 Teichnillcr space as a hafl of the cuclidean spuce
9.3 Straight lines in Tcichrniillcr space
9.4 Composition of Teichmller mappings
9.5 TeichmUller discs.
9.6 Complex structure aad feiclunlier metric
9.7 Surfaces of finite type
Bibliography
Index
206
207
208
209
211
212
212
214
215
216
218
219
220
221
221
222
223
224
226
227
229
231
232
?32
33
235
236
237
240
240
241
242
243
244
2'15
247
IntroductiOn
I.
theory of Teicbmller spaces studies the different conformal structures

on a
After the iittroduction of quasiconformal mappings
into the
the theory can be said to deal with classes consisting of
quasionfonnal mappings of a Riemann surface which are homotopic
'Fhe
modulo conformal mappings.

It was Teichmller who noticed the deep connection between quasicon-
formal mappings and function theory. He also discovered that the theory
of Tcichmller spaces is intimately connected with quadratic diffei1entials.
Teichmller ([1], {2]) proved that on a compact Riemann surfac, of genus
greater than one, every holomorphic quadratic differential
a quasiconformal mapping which is a unique extremal in its homotopy class in the
sense that it has the smallest deviation from conformal mappings. He also
showed that all extremals are obtained in this manner. It follows that the
Teichmller space of a compact Riemanu surface of genus p> I is hmemorphic to the euclidean space tR66.
TeichmUller's proofs, often sketchy and
with conjectures, were
by Ahifors
also introduced a more flexible definiput ona
paper of Ahifors revived interest in
tion for quasiconfOrmal mappings.
study of the general
Teichmiiller's work and gave rise to a
of quasiconfonnalmppings iii the plane.

Another approach to the Teichmfihler theory, initiated by Bers in the any
an entirely different manier. :Thjs
sixties, leads to quadratic differentials
in that it can also beapplied to non-cdfhpa&
method is more
now &hwartian
of
mann surfacest The quadratic differentials
of quasicohformal thappings considered on the uMconfdrmal
the extensions being obtained by use of the
versal covering
differential
Introduction
The development of the theory of Teichmller spaces along these lines

gives rise to several interesting problems which belong to the classical theory
of univalent analytic functions. Conseqoently, in the early seventies' a special
branch of the theory of univalent functions, often studied without any connections to Riemann surfaces, began to take shape.
The interplay between the theory of univalent functions and the theory
of Teichmuller spaces is the maip theme of this monograph. We do give a
proof of the above mentioned classical uniqueness and existence theorems of
Teichmller and discuss their consequences. But the emphasis is on the study
of the repercussions of Bers's method, with attention both to univalent functions and to Teichmller spaces. It follows that even though the topics dealt
with provide an introduction to the Teichmller theory, they leave aside

many of its important aspects. AbikofT's monograph [2] and the surveys of
Bers [10], [I 1], Earle [2], Royden [2], and Kra [2] cover material on Teichmuller spaces not treated here, and the more algebraic and differential geometric approaches, studied by Grothendieck, Bets, Earle, Thurston and many
others, are not considered.
There is no clearly best way to organize our material. A lot of background
knowledge is needed from the theory of quasiconformal mappings and of

Riemann surfaces. A particular difficulty is caused by the fact that the interaction between univalent functions and Teichmlier spaces works in both
directions.
Chapter I is devoted to an exposition of quasiconformal. mappings. We

have tried to collect here all the basic results that will be needed later. For
detailed proofs we )lsually refer to the monograph LehtoVirtanen [1]. The
exceptions are cases where a brief proof can be easily presented or where we
have preferred to use different arguments or, of course, where no precise
reference can be given.
Chapter 11 deals with problems of univalent functions which have their

origin in the TeichmUller theory. The leading theme is the interrelation between the Schwarzian derivative of an analytic function and the complex
dilatation of its quasicoforinal extension. A large fraction of the results of
Chapter II comes into direct use in Chapter III concerning the universal
Teichmller space. This largest and, in many ways, simplest Teichmller
space links univalent analytic functions dnd general Teichmller spaces.
A presentation of the material contained in Chapters II and UI paralleling
the introduction of the Teichniller space of an arbitrary Riemann surface
would perhaps have provided a better motivation for some definitions and
theorems in these two chapters. But we hope that the arrangement chosen
makes the theory of Teichmihler spaces of Riemann surfaces in Chapter V
more transparent, as the required hard analysis has by then largely been dealt
clear division of the book into two parts: Chapters 1,
with. Also, we
II and III concern complex analysis in the plane and form an independent
entity even without the rest of the book, while Chapters IV and V are related
to Riemann surfaces.
Introduction
The philosophy of Chapter IV on Riemann surfaces is much the same as

that of Chapter I. The results needed later are formulated, and for proofs
references are usually made to the standard monographs of AhlforsSario
[1], Lehner [1], and Springer [1]. An exception is the rather extensive treatment of holomorphic quadratic differentials, which are needed in the proof of
Teichmller's uniqueness theorem. Here we have largely utilized Strebel's
monograph [6].
Finally, after all the preparations In Chapters IIV, Teichmller spaces of
Riemann surfaces are taken up in Chapter V. We first discuss their various
characterizations and, guided by the results of Chapter III, develop their

general theory. After this, special attention is paid to Teichmller spaces of
compact surfaces. The torus is first treated separately and then, via the study
of extremal quasiconformal mappings, compact surfaces of higher genus are
discussed.
Each chapter begins with an introduction which gives a summary of its
contents. The chapters are divided into sections which consist of numbered
subsections. The references, such as 1.2.3, are made with respect to this threefold division. In references within a chapter, the first number is omitted.
In this book, the approach to the theory of Teichmller spaces is based on
classical complex analysis. We expect the reader to be familiar with the

theory of analytic functions at the level of, say, Ahlfors's standard textbook
"Complex Analysis". Some basic notions of general topology, measure and
integration theory andfunctional analysis are also used without explanations. Some acquaintance with quasiconformal mappings and Riemann sur-
faces would be helpful, but is not meant to be a necessary condition for

comprehending the text.
CHAPTER 1
Intioduction to Chapter I
Quasiconformal mappings are an essential part of the contents of this book.
They appear in basic definitions and theorems, and serve as a tool over and
over again.
Sections 14 of Chapter I aim at giving the reader a quick survey of the
main features of the theory of quasiconformal mappings n the plane. Complete proofs are usually omitted. For the details, an effort was made to give
precise references to the literature, in most cases to the monograph Lehto
Virtanen [11.
Section 1 introduces certain conformal invariants. The Poincar metric is
repeatedly used later, and conformal modules of path families appear in the
characterizations of quasiconformality.
In section 2, quasiconformality is defined by means of the maximal dilatation of a homeomorphism. Certain compactness and distortion theorems,
closely related to this definition, are considered. Section 3 starts with the
classical definition of quasiconformal diffeomorphisms and explains the con'nections between various
arid analytic properties of quasiconformal mappings.
Section 4 is concerned with the characterization of quasiconforrnal mapequations. Comas homeomorphic. solutions of Reltrami
plex dilatation, a central notion throughout our presentation, is introduced,
and the basic theorems about the existence, uniqueness and representation of
a quasiconformal mapping with prescribed complex dilatation are discussed.
than sections 1 --4, and
The rcrnainiftg two sections are more
their contents are more clearly determined by subsequent applications. Section 5 is devoted to the now classical problem of extending a homeomorphic
_____
I. Conformal Invanants
self-mapping of the halfself-mapping of the real axis to a

plane. The solution is used later in several contexts.
complex dilaistion and
Section. 6 deals with quasidisca. Along with
the Schwarzian derivative the notion of a quasidisc is a. trademark of this
book. For this reason, we have given a fairly comprehensive account of their'
numerous geometric properties, in most cases with detailed proofs.
'1. ConfOrmal Invar ants

1.1. Hyperbo!ic Metric
primarily
In the firstthree chaptets of this monograph, we shall be
with mappings whose domain and rang are subsets of the

Unless
otherwise stated, we understand by "plane" the Riemann sphere and often
use the spherical metric to remove the special position of the point at infinity.
In addition to the euclidean and spherical metrics, we shalt repeatedly

avail
of a conformally invariant hyperbolic metric. In
"unit
disc D = {zl 121 < 1) one arrives at this metric by considering Mbius transformations z
w,
16
ZZ0 ..,
which map D onto itseli By

there are no other conformal
self-mappings of 15. It follows that the differential
:
Fdzl
1_1z12
undet the group of conforthal mapings of
defines a'metric which is

D onto itself.
The shortest curve in this metric joining two points and 22 of D is the
The hyperbolic distance
circular arc which is orthogonal to the unit
between z1 and z2 is given by the formula
1
I'1
Z21
(Li)
,.
The Riemann mapping theorem says that every simply connected
A of the plane with more than one boundary point'is conformally equivalent
to the unit disc. Let f: 'A D be a-conformal mapping, and
,,Az)
If'(z)l
Then the differential

71A(Z)IdZl
Mappuigi
1.
which is
defines the hyperbolic (or Poincar) metric of A. The function
called the Poincar density of A, is well defined, for it does not depend on the
particular choice of the mappingf. In the upper half-plane,
= 1/(2 Im z).
The geodesics. which are preserved under conformal mappings, are called
hyperbolic segments.
The Poincar density is monotonic with respect to the domain: If A1 is a
simply connected subdomain of A and z e A1, then
(1.2)
q4,(z).
For let f and 11 be conformal maps of A and A1 onto the unit disc D, both
and application of Schvanishing at z. Then q4(z) = (f'(z)I,
=
warz's lemma to the function fofr1 yieldi(1.2).
Similar reasoning gives an upper bound for
in terms of the eucidean
distance d(z,
from z to the boundary of A. Now we apply Schwarz's
lemma to the function 0 f(z + d(z, A)C) and obtain
d(z,A)
(1.3)
For domains A not containing oo we also have the lower bound

(1:4)
This is proved by means of the Koebe one-quarter theorem (Nehari [2],

p. 214): If f is a conformal mapping of the unit disc D with f(O) = 0, f'(O) =
1, f(z)
then d(O,8f(D)) 1/4. We apply this to the function w
(g(w) z)/g'(O), where g is a conformal mapping of I) onto A with g(0) = z.
Because
= 1/ Ig'(O)I, the inequality (1.4) follows. Both estimates (1.3) and
(1.4) are sharp.
There is another lower estimate for the Poincar density which we shall
need later. Let A be a simply connected domain and w1, w2 finite points
outside A. Then
vw7j
(1.5)
for every zeA. To prove (1.5) we observe that z .f(z) =(z w1)/(z
maps A onto a domain A' which does not contain 0 or
Hence,
w2)
by the
conformal invariance of the hyperbolic metric and by (1.4),

fIA(Z)
'IA.(f(Z)) If(z)I
4d(f(z), A')
Since d(f(z),8A') If(z)I, the inequality (1.5) follows.

The hyperbolic metric can be transferred by means of conformal mappings
to multiply connected plane domains with more than two boundary points
and even to most Riemann surfaces. This will be explained in JV.3.6. Finally,
1. Conformal Invariants
in V.9.6 we define the hyperbolic metric on an arbitrary complex analytic

manifold.
1.2. Module of a Quadrilateral

A central theme in what follows is to measure in quantitative terms the
deviation of a homeomorphism from a conformal mapping. A natural
way to do this is to study the change of some conformal invariant under
homeomorphisms.
In 1.6 we shall exhibit a general method to produce conformal invanants
which are appropriate for this purpose. Hyperbolic distance is not well suited
to this objective, whereas two other special invariants, the module of a quad-
rilateral and that of a ring domain, have turned out to be particularly important. We shall first discuss the case of a quadrilateral.
A Jordan curve is the image of a circle under a homeomorphism of the
plane. A domain whose boundary is a Jordan curve is called a Jordan domain.
Let f be a conformal mapping of a disc D onto a domain A. Suppose that
A iS locally connected at every point z of its boundary , i.e., that every
neighborhood U o z in the plane contains a neighborhood V of z, such that
V A is connectea. Under this topological condition on A, a standard lengtharea argument yields the important result thatf can be extended to a homeomorphism between the closures of D and A. It follows, in partiular, that 0A
is a Jordan curve (Newman f1J, p. 173).
Conversely, a Jordan domain is locally connected at every boundary point.
We conclude, that a conformal mapping of a Jordan domain onto another
Jordan domain has a homeomorphic extension to the boundary, and hence
to the whole plane. For such a mapping, the iniages of three boundary points
can, modulo orientation, be prescribed arbitrarily on the boundary of the

image domain. In contrast, four points on the boundary of a Jordan domain
determine a conformal module, an observatin we shall now make precise.
A quadrilateral Q(z1,z2,z3,z4) is a Joidan domain and a sequence of four
points z1, z2, z3, Z4 on the boundary Q following each other so as to
determine a positive orientation of 0Q with respect to Q. The arcs (z1,z2),
(z2,z3), (z3,z4) and (z4,z1) are called the sides of the quadrilateral
-
Let f be a conformal mapping of Q onto a eucidean rectangle R. If the
boundary correspondence is such that f maps the four distinguished points

z1, z2,z3, z4 to the vertices of R, then the mapping f is said to be canonical,
and R is called a canonical rectangle of Q(z1,z2,z3,z4). Itie not difficult to
prove that every quadrilateral possesses a canonical mapping and that the
canonical mapping Is uniquely determined up to similarity transformations.
The existence can be shown if we first map Q conformally onto the upper
half-plane, arrange the four distinguished points in pairwise symmetric positions with respect to the origin, and finally perform a conformal mapping by
means of a suitable elliptic integral. The uniqueness part follows directly from
_____
I.
the reflection principle. (For the details, see LehtoVirtanen [1]; for this
monograph, ro which several references will be made in Chapter 1, we thall

henceforth use the abbreviation
<x <a,O <y <b} is a canonical rectNow suppose that R = (x +
z2) corresponds to the line
angle of Q(z1 , z2, 23, z4) and that the first
segment 0 x a. The number a/b, which does not depend on the partithe (conformal) module of the
cular choice of the canonical rectangle, is
z2, z3, 24). We shall use the notation
quadrilateral
M(Q(z1,z2,z3,z4)) = a/b
for thc module. It follows from the definition that M(Q(z1 ,22, z3,
1/M(Q(z2, z3, z4, Zi))..
From the definition it is also clear that the module of a quadrilateral is
conformally invariant, i.e., if f is a conformal mapping of a domain A and

Q(z1,z2,z3,z4) is a quadrilateral such that Q A and 1(Q) is a Jordan domain, then M(Q(z1, z2,23,z4)) = M(f(Q)(f(z1),f(z2),f(z3),f(z4))).
1.3. Length-Area Method.

It is possible to arrive at the notion of the module of a quadrilateral through
an extremal problem, by use of a length-area method. This approach has
turned out to be extremely useful and it leads to far-reaching generalizations,
even beyond complex analysis. In the general situation we shall discuss it in
1.6. I.n explicit form the idea was announced by A. Beurling in the 1946
Scandinavian Congress of Mathematicians in Copenhagen, and a few years
later it was used systematically for the first time by L. Ahifors and A. Beurlin&
In order to arrive at this characterization of the module we consider the

canonical mapping f of the quadrilateral Q(z1,z2,z3,z4) onto the rectangle
R={u+ivfO<u<a,0<v<b}.Then
If
J.JQ
Let r be the family of all locally rectifiable Jordan arcs in Q which join the
sides (z1 , z2) and (z3, z4). Then
I If'(z)IIdzI
for every
y is the inverse image of a vertical line segment
of R joining its horizontal sides. Hence
M(Q(z11z2,z3,z4))=
JJ
yer
(1.6)
Conformal lavariante
We can get rid of the canonical mapping f if we introduce the family P

whose elements p arc non-negative Borel-measurable functions in Q and
satisfy the condition L p(z)ldzI I for every y eF. With the notation
m,(Q)=J$p2dxdY,
we
then have
M(Q(z1,z2,z3,z4)) = inf
(1.7)
PEP
This basic formula can be proved by a length-area reasoning. Define for every
of) Lt'I = p.
given p e P a function Pt in the canonical rectangle R by

Then, by Fubini's theorem and Schwarz's inequality,
IL
du(J' Pi(U + iv)dv).
The last integral at right is taken over a line segment whose preimage is in r.
Therefore, the integral is 1, and so
a/b M(Q(z1,z2,z3,z4)). To
complete the proof we note that p If'I/b belongs, to P. By (1.6) this is an
= M(Q(z1, z2, z3, z4)).
extremal function for which
1.4. Rengel's Inequality

The power of the characterization (1.7) is that it yields automatically upper
estimates: M(Q(z1,z2,z3,z4)) m,(Q) for any pEP. An important application is obtained if we choose p to be the euclidean metric. Let denote the
euclidean distance of the sides
z2) and (z3, z4) in Q, and m the euclidean
area. Then (1.7) gives Rengel's inequality
M(Q(z1,z2,z3,z4))
(1.8)
It is not difficult to prove that equality holds if and only if Q(z1, z2, 23, z4) is
a rectangle with its usual vertices ([LV), p. 22).
Using (1.8) we can easily prove that the module depends continuously on
the quadrilateral. For a precise formulation of the result let us consider a

sequence of quadrilaterals
n
1, 2,.... Suppose that this
sequence converges to Q(z1,z2,z3,z4) from inside, i.e.,
Q for every n and
to every r > 0 there corresponds an ; such that for n

every point of the
sides of
has a spherical distance <e from the corresponding
sides of Q(z1,z2,z3,z4). Then
Jim
zfl)) = M(Q(z1, 22, Z3, Z4)).
To prove this, we only need to carry out a canonical mapping of Q(z1,z2,

zr).
z3, z4) and apply Rengel's inequality to the image of
L Quasiconfonnal Mappings
10
Rengel's inequality also makes it possible to characterize conformality in

terms of the modules of quadrilaterals, without any a priori differentiability.
Theorem 1.1. Let f:
A ' A' be a sense-preserving homeomorphism which leaves

invariant the modules of the quadrilaterals of the domain A. Then f is conformal.
sketch a proof. Map a quadrilateral of A and its image in A' canonically

Onto identical rectangles R and R' whose sides are parallel to the coordinate
axes. Given a point z = x + ly of R, we consider the two rectangles R1 and
R2 onto which R is divided by the vertical line through z. Since all modules
remain invariant, it follows from Rengel's inequality, with regard to the

possibility for equality, that the images of R1 and R2 in R' are also rectangles
(cf. [LV), p. 29). But then the real part of the image of z must be x. A similar
argument shows that the
part of z does not change either. Thus the
induced mapping of R onto R' is the identity, and the conformality of f

follows.
Module of a Ring Domain

A doubly connected domain in the extended plane is called a ring domain.
Unlike simply connected domains, which fall into three conformal equivalence classes, ring domains possess infinitely many
equivalence
classes. A counterpart for Riemann's mapping theorem says that a ring

domain can always be mapped conformally onto an annulus r < jzI <R,
where r 0, R 00. it follows that every ring domain B is conforinally
< 00,30
equivalent to one of the following annuli: 100 <
<
I
<R, R < 00. In case 30 the number R determines the equivalence
class, and
M(B)
log R
is called the module of B. In cases 1 and 2 the module of B is said to be

infinite. A conformal mapping of B onto an annulus is called a canonical
mapping of B.
Just as in the case of quadrilaterals, the module of ring domains can also
r now be the family
be deflated without reference to canonical mappings.
Jordan curves in a ring domain B which separate the boundof
As before, P is the family of all non-negative Borelmeasurable functions in B with
I for every ye!'. Then
M(B) = 2n inf
,cP
By use of this formula, many geometrically more or less obvious statements can be rigorously founded. We list here some applications. The first
result says that the module cannot be large if neither of the boundary corn-
1. Conformal Invanants
11
ponents is small in the spherical metric: Let B be a ring domain which separates the points a1, b1 from the points a2, b2. If the spherical distance between
a and
i=
1,
2, is b, then
M(B) ,t2/2c52
(1.9)
The second estimate shows that if the boundary components are close to
each other and none of them is small, then the module is small. More precisely: Let B be a ring domain whose boundary components have spherical
diameters >5 and a mutual spherical distance <e <5. Then
M(B) n2/log
(1.10)
For the proofs of(1.9) and (1.10), see [LV], p. 34.

The third inequality solves an extremal problem. We introduce the Grtzsch
ring domain whose boundary components are the unit circle and the line
denote its module. If B is a ring
segment {xIO x r}, 0 < r < 1; let
domain separating the unit circle from the points 0 and r, then
M(B)
This was proved by Grtzsch in 1928 ([LV], p. 54).
A simple application of the reflection principle shows that the Teichmller
x 0 and x r2 has the
ring domain B bounded by the line segments -module
M(B) =
(1.11)
+ r2))'12).
This domain is also connected with an extremal problem: if the ring domain
B separates the points 0 and z1 from the points z2 and
then
M(B) 2,z((1z21/(1z1 I +
(1.12)
Inequality (1.12) generalizes a result of Teichmuller; for the proof we refer

to [LV], p. 56.
1.6. Module of a Path Family

We showed above that the modules of quadrilaterals and ring domains can
be defined with the aid of certain path families. We shall now consider the
more general situation in which an arbitrary family of paths is given.
By our terminology, a path is a continuous mapping of an interval into the
plane and a curve the image of the interval under a path. We feel free not to
make a very clear distinction between a path and a curve, if there is no fear
of confusion, e.g., to use the same symbol for a path and its image.
Let A be a domain and r a family of paths in A. We associate with F the
class P of non-negative Borel measurable functions p in A which satisfy the
condition
I. Quasiconformat Mappings
12
for every locally rectifiable)' in r; such a p is said to be admissible for F. The

numbcr
M(fl= inf
pEPj JA
p2dxdy
is called the module of the path family F.

module of a quadrilateral and of a ring domain are special cases of this
notion, whose properties are studied in [LV], pp. 132- 136.
isa
{foylyer}. Itfollows
from the definition of the module that if f is conformal, then M(f(F)) =
M(r), i.e., the module of a path family is conformally invariatu.
11.1: A VA'
2. Geometric Definition of Quasiconformal

Mappings
2.1. Definitions of Quasiconformality
Given a domain A, consider all quadrilaterals Q(z1,z2,z3,z4) with Q c A.
Let J': A
A' be a sense-preserving homeomorphism. The number
M(f(Q)(f(z1 ),f(z2),f(z3),f(z4)))
M(Q(z1,z2,z3,z4))
is called the maximal dilatation of f. It is always 1, because the modules of

Q(z1,z2,z3,z4) and Q(z2,z3,z4,z1)are reciprocals.
Since the module is a conformal invariant, the maximal dilatation of a

conformal mapping is 1. By Theorem 1.1, the converse is also true: if the
maximal dilatation of f is equal to I, then is conformal. From this observation we arrive conveniently at the notion of quasiconformality.
Definition. A sense-preserving homeomorphism with a finite maximal dilatation is quasiconformal. If the maximal dilatation is bounded by a number K,
the mapping is said to be K-quasiconformal.
This "geometric" definition of quasiconformality was suggested by Pfluger
[1] in 1951, and its first systematic use was by Ahlfors [1] in 1953.
By this terminology, f is I -quasiconformal if and only if I is conformal. If
f is K-quasiconformal, then M(f(Q)) M(Q)/K for every quadrilateral in

A. A mapping f and its inverse f are simultaneously
From the definition it also follows that if f: A B is K1-quasiconformal and
g: B
C is K2-quasiconformal, then g of is K1 K2-quasiconformal.
13
The definition of quasiconformality could equally well have been given in

terms of the modules of ring domains: A sense-preserving
f of
a domain A is K-quasiconformal if and only ;f the module condiion
M(f(B)) KM(B)
(2.1)
holds for every ring domain B, B c A.
The necessity of the condition can be established easily if the canonical

annulus of B, cut along a line segment joining the boundary components, is
of the logarithm. To prove the suftransformed to a rectangle with the
ficiency requires somewhat more el&$rate module estimations ([LV], p. 39).
mapping cannot "blow up"
Inequality (2.1) shows that a
a point, and the well known
on the removability of isolated singularities of conformal mappings
be readily generalized ([LV), p. 41):
A K-quasiconforma! mapping of a ;kiomain A with an isolated boundary point
a can be extended to a K-quasiconfor,nal mapping of A u {a}.
We mention here another generalized extension theorem (cf. 1/2): A quasiconformal mapping of a Jordan domain onto another Jordan domain can be
extended to a homeomorphism between the closures of the domains. This can be
proved by a modification of the proof for conformal mappings ([LV], p. 42),

or deduced directly from the corresponding result for conformal mappings
by use of Theorem 4.4 (Existence theorem for Beltrami
The modules of quadrilaterals and ring domains, which
to characterize quasiconformality, are modules of certain path families. As a matter of
fact, the following general result, proved by Visl in 1961, is true.
A K-quasiconformal mappingf of a domain A satisfies the inequality
M(f(F')) KM(fl
(2.2)
for every path family r of A.

Flersch, one of the pioneers in applying curve families to quasiconformal
mappings, asked as early as 1955 in his thesis whether (2.2) could be true for
all path families. At that time, certain "analytic" properties of quasiconformal
mappings, which the proof ([LV], p. 171) seems to require, were not yet
known. These properties will be discussed in section 3. The reason we men-
tion the result (2.2) here is that we could have taken it as a definition for
K-quasiconformality. In a way, a characterization by means of the general
relation (2.2) is more satisfactory than the definition which is based on the
special notion of the module of a quadrilateral. However, we have preferred
to use quadrilaterals, not merely for historical reasons, but also to remain
true to the presentation in the monograph [LV], to which repeated references are being made.
2.2. Normal Families of Quasiconformal Mappings

Let us consider a family whose elements are mappings of a plane domain A
into the plane. Such a family is said to be normal if every sequence of its
14
1.
Quasiconfonnal Mappings
elements contains a subsequence which is locally uniformly convergent in A.

To cover the possibility that x lies in A or in the range of the mappings, we
is not there, we can of course switch to the
use the spherical metric. If
topologically equivalent euclidean metric.
If a family is equicontinuous, then it is normal. This is the result on which
the proofs for a family to be normal are usually based in complex analysis.
For instance, if the family consists of uniformly bounded analytic functions,
equicontinuity follows immediately from Cauchy's integral formula, and so
normality can be deduced. A generalization of this result, proved by use of
the elliptic modular function, says that if the functions are meromorphic and
omit the same three values, then the family is normal.
Much less is needed for normality if the functions are assumed to be

injective.
Lemma 2.1. Let F be a family of K-quasiccnformal mappings of a domain A.

If every f F omits two values which have a mutual spherical distance d > 0,
then F is equiconrinuous in A.
Let s denote the spherical distance. Given an s, 0 < e < d, and

a point z0eA, we consider a ring domain B = {zI <s(z,z0) < r} with
> 0 so small that M(B)> ic2K/2g2. Let z1
{zls(z,z0) <r} c A, and
be an arbitrary point in the neighborhood V = {zls(z,z0) <5) of z0.
Paoov.
Let us consider an f F. By assumption, f omits two values a and b with

s(a, b) d. The ring domain f(B) separates the points f(z0), f(z1) from the
points a. b. If
min(d,s(J(z0),f(z1))), it follows from formula (1.9) that
M(f(B))
This yields
and hence the desired estimate s(f(z0),
V, for every fE F.
0
f(z)) <c whenever z
Lemma 2.1 yields various criterions for a family to be equicontinuous and

hence normal. The following will come into use several times.
Theorem 2.1. A family F of K -quasiconformal mappings of a domain A is

equicontinuous and nonnal, for three fixed points z1, z2, z3 of A and for
every fe F, the distances
are uniformly bounded away from zero
for i,j=
PRooF. By Lemma 2.1, the family F is equicontinuous in
1, 2, 3, i # j, and hence throughout 4.
i, j =
2.3. Compactness of Quasiconformal Mappings

be a sequence of K-quasiconformal mappings of a domain A which
is locally uniformly convergent in A. If the limit function f is not constant, it
must take at least three different values, because it is continuous. It follows
from Theorem 2.1 that the functions constitute an equicontinuous family.
Let
15
K-quasiconformal mappings possess the same compactness property as

conformal mappings.
Theorem 2.2. The limit function f of a sequence

of K-quasiconformal
mappings of a domain A, locally uniformly convergent in A, is either a constant or a K-quasiconformal mapping.
1ff is a homeomorphism, then it follows easily from the definition of

K-quasiconformality and from the continuity of the module of quadrilaterals
that f is K-quasiconformal ([LV], p. 29). A continuous injective map of an
PROOF.
open set of the plane into the plane is a homeomorphism (Newman [1],
p. 122). Therefore, it is sufficient to show that a non-constant limit function
f is injective. This we can prove, utilizing the fact that the family
is
equicontinuous, with the aid of the module estimate (1.10) ([LV3, p. 74). 0
In case every f maps A onto a fixed domain A', more can be
said
about
the limit function.
Theorem 2.3. Let A be a domain with at least two boundary points and
(fm) a sequence of K-quasiconformal mappings of A onto a fixed domain
A'. If the sequence (fN) converges in A, then the limit function is either a
K-quasiconformal mapping of A onto A', or a mapping of A onto a boundary
point of A'.
Here we need not assume that
is locally uniformly convergent, because
we conclude from Lemma 2.1 that
is a normal family. The theorem
follows from equicontinuity and normal family arguments ([LV], p. 78).

In 4.6 we shall study the convergence of K-quasiconformal mappings f,,
more closely. It turns out that, even though the mappings f, tend uniformly
towards a K-quasiconformal limit f, the local mapping properties off and

the approximating functions f,, may be quite different.
2.4. A Distortion Function

In later applications we shall often encounter a distortion function which we
shall now introduce, starting from its simple geometric interpretation.
Let F be the family of K-quasiconformal mappings of the plane which map
the real axis onto itself and fix the points 1,0 and x. By Theorem 2.1, F is
a normal family, and so
= max{f(1)jfeF}
exists. This defines our distortion function
more explicit expression.
Consider the quadrilateral H( 1,0,1,
(2.3)
for which we shall now derive a

where H is the upper half-plane.
L Quasiconformal Mappings
16
The Mbius transformation z .(1 + z)/(1 z) maps it onto the quadri 1). Hence, these two quadrilaterals have the same module.
1,
hand, the modules are reciprocal, and so M(H( 1,0,1, co)) =
On the
M(H(0, 1,c4 1))= 1.
Let us
an feF and write t = f(1). We form the Teichmuller ring B
by the line segments 1 x 0 and x t. If B is conformally
equivalent to the annulus A = (ill <IzI <R}, then the canonical mapping
of B can be so chosen that the upper half of B is .conformally equivalent to
upper hail of A. By applying the mapping z ' log z we conclude that
M(H(0, t,
1))
M(B)
M(H(0, t,
1)) KM(H(O, 1, co,
mates, we obtain
1))
=
f is K-quasiconformal,
= K. if we combine all these esti-
t=f(1)(1r1(irK/2))21.
(2.4)
In order to show that there is an f for which equality holds, we fIrst make
a general remark: Let f be a homeomorphism of a domain A and I a closed
line segment which lies in A with the possible exception of its endpoints. Then
f has the same maxima! dilatation A and, in A\I. This can be proved by
means of Rengel's inequality ([LV], p. 45), or by making use of the analytic
characterization of quasiconformality which will be given in 3.5. It follows
that the reflection principle for conformal mappings generalizes as such for
K-quasiconfonnal mappings. In particular, a K-quasiconformal self-mapping
of the upper half-plane can be extended by reflection in the real axis to a
K-quasiconformal mapping of the plane.
Let us now return to (2.4). Let 11 be the canonical mapping of the quadrilateral H(0, 1,00, 1) onto the square Q(O, 1,1 + i,i), the affine stretching
x + iy Kx + iy, and 12 the canonical mapping of H(0, t, 00, 1) onto the
o ofj
rectangle R(O, K, K + i, i). Then the mapping f which is equal to
in H
its mirror iiage in the lower hall-plane is K-quasiconforinal in the
plane. For this f, equality holds in (2.4). It follows that
2(K) =
From the obvious result 2(1) =
1.
1
we conclude that
n/2.
(2.5)
Typically the reasoning goes in the other direction, in that we retrieve information about 2(K) by estimating p(r). For instance, we obtain in this way
2(K) =
+ o(l),
with a positive remainder term o(1) as K
00 ([LV]. p. 82). Also,
1(K) exp(4.39(K
1))
(BeurlingAhifors [1]), which tells about the behavior of 2(K) as K -+
1.
In
17
particular, we see that A. is continuous at K

follows from the continuity of r , p(r).
1; continuity at an arbitrary K
2.5.
Circular Distortion
maps the family

A conformal mapping of the plane which fixes 0 and
Under normalized K-quasiconof circles centered at the ongin onto
formal mappings the images of these circles have "bounded distortion". To
be precise:
Theorem 2.4.. Let f be a K.quasiconf"ormal mapping of the plane fixing 0 and

ri.. Then for every r> 0,
min,jf(re")I
c(K),
(2.7)
where the constant c(K) depends only on K.
There are many ways to prove this important theorem. A normal family
argument shows that a finite bound c(K) must exist. A quantitative estiand z be points on the circle
mate is obtained as follows. Let
=
r at which the minimum and maximum of lf(z)( are attained. For B' =
{wlmin,If(re1')I <Iwl <max,If(re")I}, let B be the inverse image of B'.
Then B separates the points 0, z1 from the points z2, cO. Hence, by (1.12)
I
and (2.5),
M(B) 2,t((1z11/(Iz11 + 1221))') =

Consequently,
ic.
M(B') KM(B) irK, and it follows that (2.7) holds for
c(K) =
in
The sharp bound in (2.7) is A(K) (proved by
1959). This follows from the fact that, as a generalization of (2.3), ).(K) is the
maximum of tf(z)I on the unit circle in the family of K-quasiconformal

mappings of the plane which fix 1,0 and oo but which are not required to
map the real axis onto itself. There seems to be no easy way to prove this'
result.
Theorem 2.4, in a form in which the value of the sharp bound is not
needed, will render us valuable Service in section 6 when we study the geometry of quasidiscs. With these applications in irnad, we draw here a further
conclusion from (2.7).
Letf be a K-quasiconformal mapping of the planlxing

(2.7) that if Iz2 zol Izj
201,
We infer from
then
lf(z2)f(z0)I c(K)If(z1)f(zo)I.
The following generalization is also readily obtained. If 122
where n 1 is an integer, then
(2.8)
201
nIz1
zol,
1. Quasiconformal Mappings
18
If(z2)f(zo)l nc(Krlf(za)f(zo)I;
k = 0, 1,
In order to prove this, we denote by
points onthe ray from z0 through z2 for which
=
= Z2. By (2.8),
2,..., n. Hence, by the triangle inequality,
If(Z2)
By(2.8), 11(C1)
(2.9)
...,
I
n, the equidistant
1z2 z01/n; here
f(C12)l fork =
f(zo)IEc(K)k
f(zo)I c(K)lf(z1)
f(z0)1.
and (2.9) follows.
We conclude this section with the remark that quasiconformality can be

defined by means of the distortion function H,
H(z) = hmsup
mai.lf(z +
.
mm,If(z
+ re')
though we shall not make use of this characterization. A sense-
r-'O
even
preserving homeomorphism f of a domain A is .K-quasiconformal if and only
if H is bounded in
and H(z) K almost everywhere in A
([LV], pp. 177178).
3. Analytic Definition of Quasiconfrmal

Mappings
3.1. Dilatation Quotient
When the definition of quasiconformality in terms of the modules of quadrilaterals was given in the early fifties, quasiconformal mappings had been
studied and successfully applied in complex analysis for more than two
decades. Historically, the starting point for generalizing conformal mappings
was to consider, not arbitrary sense-preserving homeomorphisms, but dtffeomorphisms, i.e., homeomorphisms which with their inverses are continuously
differentiable. We can then generalize the characteristic property of conformal mappings that the derivative is independent of the direction by requiring
that the ratio of the maximum and minimum of the absolute value of the
directed derivatives at a point is uniformly bounded.
We shall now show that this classical definition gives precisely those quasi-
conformal mappings which are diffeomorphic. This local approach using

derivatives is often much more convenient than the definition using modules
of quadrilaterals when the problem is checking the
of a
mapping given by an analytic expression.
19
3. AnaJytic Definition of Quasiconformal Mappings
To make the above remarks precise, we introduce for a sense-preserving

diffeomorphism f the complex derivatives
and the derivative 8j in the direction
+
Then
rem)
f(z)
and so
=
max fj(z)( =
mm IOJ(z)l =
is positive, because the Jacobian J1 = 18112

The difference IOf(z)I
is
positive
for
a
sense-preserving
diffeomorphism. We conclude that the
Of
I
dilatation quotient
D max1I8jI
is finite.
The mapping f is conformal if and only if Of vanishes identically. Then Oj
is independent of we have Of Of = f'. This is equivalent to the dilatation quotient being identically equal to 1.
The dilatation quotient is conformally invariant If g and h are conformal
such that w = h of o g is defined, then direct computation shows
that D,(z) =
3.2.
Quasiconformal Diffeomorphisms
For diffeomorphisms quasiconformality can be characterized with the aid of

the dilatation quotient.
Theorem 3.1. Let I: A , A' be a sense-preserving diffeomorphism with the
property
D1(z) K
for every z A. Then f is a K-quasiconformal mapping.
PROOF. We pick an arbitrary quadrilateral Q of A. Let w be the mapping

which is induced from the canonical rectangle R(O, M, M + i, i) Of Q onto
the canonical rectangle R'(O, M', M' + I, i) of 1(Q). Because of the conformal
invariance of the dilatation quotient,
is also majorized by K. Hence
KJ,,, and the desired result M' KM follows by use

of a customary length-area reasoning;
20
I. Quaszconformal Mappings
M' = m(R')
Theorem 3.1 is the classical definition of K-quasiconformality given by

Grtzsch [1] in 1928.
The converse to Theorem 3.! is as follows:
Theorem 3.2. Let f: A
A' be a K-quasiconformal mapping. If f is differenti-
able at z0eA,then
max
K mm
(3.1)
The idea of the proof is to consider a small square Q centered at z0 and

regard it as a quadrilateral with the vertices at distinguished points. The area
and the distance of the sides of 1(Q) can be approximated by expressions
involving the partial derivatives of f at z0. Application of Rengel's inequality
then yields a lower estimate for M(f(Q)) from which the desired inequality
(3.1) follows. (For details we refer to [LV], p. 50.)
By combining Theorems 3.1 and 3.2 we obtain the following characterization for quasiconformal diffeomorphisms: A sense-preserving diffeomorphism
f is K-quasiconformal if and only if the dilatation condition D1(z) K holds
everywhere.
The class of K-quasiconformal diffeomorphisms does not possess the com-
pactness property of Theorem 2.2. This is one of the reasons for replacing the
classical definition of Grtzsch by the more general one. Another reason will
be discussed in 4.5.
3.3. Absolute Continuity and Differentiability

We shall soon see that an arbitrary quasiconformal mapping of a domain A
is differentiable almost everywhere in A. From Theorem 3.2 it then follows
that the dilatation condition (3.1) is true at almost all points of A. However,
the converse is not true. ii, a sense-preserving homeomorphism I which is
differentiable ac. and satisfies (3.1) a.e. is not necessarily K-quasiconformal.
What is required is a notion of absolute continuity.
A continuous real-valued Iuncon u is said to be absolutely continuous on
lines (ACL) in tdomain 4 if for each closed rectangle {x +
x b,
c y d } c A, the function x i u(x + ly) is absolutely continuous on [a, b)

for almost all ye fc, d] and y . u(x + iy) is absolutely continuous on [c, d]
for almost all xe [a, b). A complex valued function is ACL in A if its real and
imaginary parts are ACL in A.
21
3. Analytic Definition of Quasiconformal Mappings
It follows from standard theorems of real analysis that a function f which

is ACL A has finite partial derivatives and f, a.e. in A.
Theorem 3.3. A quasiconformal mapping is absolutely continuous on lines.
This result was first established by Strebel (1955) and Mon. A later proof
by Pfluger, which uses Rengel's inequality and a minimum of real analysis, is
presented in [LV], p. 162.
From Theorem 3.3 we conclude that a quasiconformal mapping has finite
partial derivatives a.e. From this we can draw further conclusions by making
use of the following result:
Let I be a complex-valued, continuous and open mapping of a plane domain

A which has finite partial derivatives a.e. in A. Then f is differentiable a.e. in A.
The proof, which is due to Gebring and Lehto (1959), uses the maximum
principleand a standard
on the density of point sets ([LV], p. 128).
Application to quasiconformal mappings yields, with regard to Theorem 3.2,
a basic result:
Theorem 3.4. A K-quasiconformal mapping f of a domain A is differentiable

and satisfies the dilatation condition (3.1) almost everywhere in A.
Differentiability a.e. of quasiconformal mappings was first proved by Mon

[I] with the aid of the RademacherStepanofr theorem and Theorem 2.4.
3.4. Generalized Derivatives
The ACL-property, which depends on the coordinate system, becomes much
more useful when combined with local integrability of the derivatives. A

function f is said to possess (generalized) V-derivatives in a domain A, p 1,
if f is ACL in A and if the partial derivatives f and f,, off are L"-integrable
locally in A. It is also customary to say that the function f then belongs to
the Sobolev space
This property is preserved under continuously
differentiable changes of coordinates ([LV], pp. 151152).
Roughly speaking, classical transformation rules of Calculus between curve
and surface integrals remain valid for functions with V-derivatives. This is
one reason for the importance of this class of functions. (For details and more
information see, e.g., [LV], pp. 143154 or Lehto [4], pp. 127131.)
A quasiconformal mapping has L2-derivatives. In order to prove this we first
note that the dilatation condition (3.1) implies the inequality
KJ(z).
In particular,
KJ(z),
< KJ(z) a.e. The Jacobian of an almost
22
everywhere differentiable homeomorphism is locally integrable ([LV],

p. 131). Consequently, f and f, are locally L2-integrable.
A homeomorphism with L2-derivinives is absolutely continuous with respect
to two-dimensional Lebesgue measure ([LV], p. 150). Thus quasiconformal
mappings have this property. They carry sets measurable with respect to
two-dimensional measure onto other sets in this class. The formula
= m(f(E))
(3.2)
JE
for every quasiconformal mapping f of a domain A and for every

measurable set E c A. If we apply (3.2) to the inverse mapping f1, we
holds
deduce that for a quasiconformal mapping J(z) > 0 almost everywhere.

The considerations in 4.4 will show that every quasiconformal mapping
has not only L2-derivatives but actually U-derivatives for some p> 2. This
is a much deeper result than the existence of L2-derivatives.
3.5. Analytic Characterization of Quasiconformality

A simple counterexample, constructed with the help of Cantor's function,
shows that a homeomorphism need not be quasiconformal even though it
is differentiable a.e., satisfies (3.1) a.e. with K = 1, has bounded partial derivatives, and is area preserving ([LV], p. 167). What is required is the
ACL-property, but once this is assumed it together with (3.1) guarantees
quasiconformality.
Theorem 3.5. A sense-preserving homeomorphism
conformal
1
f of a domain A is K-quasi-
fisACLinA;
a.e. in A.
We first note that being ACL, the mapping has partial de*'atives
a.e. and, as a homeomorphism, is therefore differentiable a.e. Thus condition
2 makes sense. As above, we conclude that f has L2'dcrivatives. After this,
we can follow the proof of Theorem 3.1, apart from obvious modifications.
PROOF.
0
Theorem 3.5 is called the analytic definition of quasiconformality. Apparently different from the equivalent geometric definition, it sheds new light on
the connection with the classical Grtzsch mappings. In the next section we
shall show that the analytic definition can be written in the form of a lifferential equation. This leads to essentially new problems and results for quasiconformal mappings.
Under the additional hypotheses that f is differentiable a.e. and has
23
4. Beltrami D erential Equation
in 1955. Later
L'-derivatives, Theorem 3.5 was first established by
Bers and Pfluger relaxed the a priori requirements. Under the above minimal conditions, the theorem was proved by Gehring and Lehto in 1959.
(For references and more details, see [LV], p. 169.)
4. Beltrami Differential Equation

4.1. Complex Dilatation
Inequality (3.1) is a basic property of quasiconformal mappings. The very
natural step to express therein more explicitly the maximum and minimum
leads to the important notion of complex dilatation and reveals a
of
connection between the theories of quasiconformal mappings and partial
differential equations.
Let 1: A -, A' be a K-quasiconformal mapping and z e A a point at which
f is differentiable. Since maxIc3jt
the
I=
dilatation condition (3.1) is equivalent to the inequality

If(z)I.
(4.1)
Suppose, in addition, that J1(z)> 0. Then af(z)

quotient
0, and we can form the
3f(z)
The function p. so defined a.e. in A, is called the complex dilatation off. Since
j is continuous, p is a Borel-measurable function, and from (4.1) we see that
Ip(z)I
<1.
(4.2)
Complex dilatation will play a very central role in our representation. It

has a simple geometric interpretation. At a point z at which p is defined, the
mapping
f(z) + af(z)g z) + af(z)(C
a non-degenerate affine transformation which maps circles centered at z

onto ellipses centered at f(z). The ratio of the major axis to the minor axis of
the image ellipses is equal to (1 +
lp(z)I). *e see that the smaller
Jp(z) is, the less the mapping f deviates from a conformal mapping at the
point z. If p(z) 0, the argument of p(z) determines the direction of maximal
stretching:
assumes its maximum when = arg p(z)/2.
is
24
4.2. Qua.siconformal Mappings and the Beltrami Equation

The definition of complex dilatation leads us to consider differential equations
(4.3)
An equation (4.3), where p is measurable and

< 1, is called a Beltrami
equation. If f is conformal, p vanishes identically, and (4.3) becomes the
Cauchy-Riemann equation f 0.
A function f is said to be an Lu-solution of (4.3) in a domain A if f has
V-derivatives and (4.3) holds a.e. in A.
is K-quasiconformal if and only if f is an
pt3f, where p satisfies (4.2) for almost all z.
Theorem 4.1. A homeomorphism f

L2-soluzion of an equation
PROOF. The necessity follows frdm Theorem 3.4 and the sufficiency from
Theorem 3.5, when we note that p)) <1 implies that f is sense-preserving.
The Beltrami equation has a long history. With a smooth coefficient p, it

was considered in the 1820's by Gauss in connection with the problem of
fjnding isothermal coordinates for a given surface (cf. IV.1.6). As early as
1938, Morrey [1] systematically studied homeomorphic L2-solutions of the
equation (4.3). But it took almost twenty years until in 1957 Bers [1] observed that these solutions are quasiconformal mappings.
In 4.5 it will become apparent that (4.3) always has homeomorphic solutions, i.e., that the complex dilatation of a quasiconformal mapping can be
prescribed almost everywhere. This is a deep result. It is much easier to
handle the question of the uniqueness of the solutions of (4.3).
Let f
and
dilatations
g be quasicorformal mappings of a domain A with complex

and
D rect computation yields the transformation formula
=
= g(z),
(4.4)
valid for almost all z e A, and hence for almost all e g(A).
Theorem 4.2 (Uniqueness Theorem). Let f and g be quasiconformal mappings

of a domain A whose complex dilatations agree a.e. in A. Then
is a
conformal mapping.
PROOF.
By (4.4), the complex dilatation of fog'
rem 3.5 we deduce that

it is conformal.
vanishes a.e. From Theois 1-quasiconforinal. Hence, by Theorem 1.1,
Conversely, iffog' is conformal, we conclude from (4.4) thatf and g have

the same complex dilatation.
25
4. Beltrami Differential Equation
4.3. Singular Integrals

The Uniqueness theorem says that a quasiconlormal mapping of the plane is
determined by its complex dilatation up to an arbitrary Mbius transformation. It follows that a suitably normalized mapping is uniquely determined by p. We shall now show that, by use of singular integrals, it is possible
to derive a formula which gives the values of a normalized' quasiconformal
mapping in terms of p.
in a domain A of the = + i,Let f be a function with
with a rectifiable boundary curve.
plane and D, D A, a Jordan
Application of Green's formula yiel 4s generalized Cauchy integral formula
([LV], p. 155)
1(z) =
!!Jj'
D.
The first term on the right, a Cauchp integral, defines an analytic

in D. We conclude, in passing, that afunction f with L'-derivatives in A is
analytic if
= C a.e. in A. The second term on the right in (4.5) is to be
understood as a Cauchy principal valup.
Suppose that f has L1-derivatives in the complex plane C and that f(z) -.0
as z co If we take D =
<R}
let R oo, the first term on the
right-hand side of (4.5) tends to zero for\ every fixed z. With the notation
Tw(z) = _1-$ f\
(4.6)
Jc\CZ
we then
obtain from (4.5)
f= \
in the complex
Assume, for a moment, that w in (4.6) belongs to class
plane, i.e., cv is infinitely many times differeMiable and has a bounded support. Straightforward computation then shows that
c3Tw = Hw
(4.8)
([LV], pp. 155157), where
Hw(z) = JJc
th7,
the integral again being defined as a Cauchy
value. The linear

operator H is called the HUbert transformation. We a\so see that
Tw=w,
\.
that the operators 3 and commute with T and H, and that To and Hco
and are analytic outside the support of cv
belong to
p. 157).
The Hubert transformation can be extended as a boun4ed operator to

I <p < co. One first proves that if cv
there exists a constant
not
26
I. Quasiconformal Mappings
depending on w, such that

(4.9)
This is called the CaldernZygmund inequality; proofs are given, apart
[1],
from the original paper by Caldern and Zygmund (1952), in
Ahlfors [53 and Stein [13. Since
is dense in L' and is complete, w' can
use (4.9) to extend the Hilbert transformation to the whole space L". Inequality (4.9) then holds for every weLt (cf.
p. 159).
p. 160).
Using (4.9) we deduce that (4.8) holds i.e. for every we U (cf.
For applications it is also important to note that the norm
IIwII,=
11
depends continuously on p (Ahifors [5], DunfordSchwartz [1]).
The special case p = 2 is much easier to handle than a general p. A rather

elemen4ary integration shows that Hilbert transforniation is an isometry in
L2 ([LV), p. 157). In particular, IIH 112 = 1.
4.4. Representation of Quasiconformal Mappings

We shall now apply the results of 4.3 to quasiconformal mappings. Let f be
a quasiconformal mapping of the plane whose complex dilatation has a
bounded support. Wishing to represent Y by means of we introduce a
normalization so that ji determines f uniquely.
We first require that f(oo) =
Near infinity, where f is conformal, we
then have 1(z) = Az + B + negative powers of z. If we set A = 1,8 = 0, then
f is uniquely determined by
In a neighborhood of cc we thus have
It follows that the partial derivatives of the function z ' f(z) z, which are
locally in L2, are L2-integrable over the plane. We_conclude from (4.7) and
from the generalized lorinula (4.8) that of = I + JIOf a.e. Since Of = p01 a.e.
we thus have
Of=p+pHOf
a.e.
(4.10)
This integral equation can be solved by the customary iteration

The Neumann series obtained converges in L2, but it also converges in U,
p> 2, if p satisfies the condition
< 1.
(4.11)
More explicitly, suppose that p(z) 0 if Izi > R, and define inductively
q,1,=p,
n2,3
(4.12)
27
4. &Itrami Differential Equation
Then
(4.13)
II
Hence, under condition (4.11),

(4.14)
and it is a function in U'.
This solution gives the desired representation formula for f(z), first established by Bojarski in 1955.
Theorem 4.3. Let f be a quasiconformal mapping of the plane whose complex dilatation p has a bounded support and which satisfies the condition
(f(z) z) = 0. Then
f(z) =
To,(z),
where is defined by (4.12). The series is absolutely and uniformly convergent

in the plane.
By (4. 14),
=
z+
p > 2, it follows from Holder's inequality that f
c, H p,
constant depends only on p and R. Therefore, by (4.13),
For
PROOF. By (4.7) we havef(z)
where the
Iii'HJ'1
(4.15)
where c, depends only on p and R. (For this crucial estimate, (4.10) must be
solved in a space V with p> 2.) We conclude from (4.15) that
and that the series on the right is absolutely and uniformly convergent.
U' locally. From Of =

We proved above that under condition (4.11),
+ HOf and (4.9) we see that the same holds for Of. It follows that the partial
derivatives of a quasiconformal mapping are locally in 1/ for same p> 2
1
([LV), p. 215). The p will, of course, depend on
4.5.
Existence Theorem
In proving Theorem 4.3 we started from a quasiconforinal mapping which

gave the function ji. The following result, fundamental in the theory of quasi.
conformal mappings, shows that we could equally well have started from a
measurable function p.
28
Theorem 4.4 (Existence Theorem). Let p be a measurable function in a domain

A with p <1. Then there is a quasiconformal mapping of A whose complex
dilatation agrees wit h'p a.e.
The proof can be divided into three parts. One first shows that if p E
the Beltrami equation
p3w has a locally irijective solution. This can be
so constructed that a topological argument shows it to be in fact globally
injective. Another way to obtain from locally injective solutions a globally
injective one is to use the general uniformization theorem for Riemann surfaces. (Theorem IV.3.3). Finally, we get a general solution by approximating
the given p with Cr-functions (cf. Theorem 4.5 below). For details of the
proof and historical remarks we refer the reader to Lebto 14], p. 136. The
proof in [LV], p. 191, employs step functions, while Vekua [1] makes use of
the explicit expression in Theorem 4.3.
For continuous p, the solutions of the Beltrami equation are not necessarily continuously differentiable. In other words, for a diffeomorphic quasiconformal mapping its complex dilatation, which is continuous, cannot be
prescribed as an arbitrary continuous function z with
< 1. This is one
more reason to generalize the classical Grtzsch definition of quasiconformality (cf. the remark made at the end of 3.2).
If p is a little more regular than just continuous, we are back in the classical
situation. For instance, a quasiconformal mapping whose complex dilatation is
locally Holder continuous is a diffeomorphism. (See [LV], p. 235, where this
conclusion is drawn from a still weaker condition on p.)
Theorem 4.4 gives immediately a striking generalizati n of the Riemann
mapping theorem: Let A and B be simply connected donu ins in the extended
plane whose boundaries consist of more than one point, and let p be a measurable function in A with p
<1. Then there is a quasiconformal mapping
of A Onto B whose complex dilatation agrees with p a.e.
In fact, by Theorem 4.4 there exists a quasiconformal mapping f of A with
complex dilatation equal to p a.e. The boundary of the simply connected

domain f(A) consists of more than one point. Hence, by Riemann's mapping
theorem, there is a conformal map g of f(A) onto B. Then g of has the desired
properties.
4.6. Convergence of Complex Dilatations
it is important in proving Theorem 4.4 that we can initially consider a

smooth p and then obtain the general result by approximation. Let us now
study more closely what relations there are between the convergence of
mappings and that of their complex ditatations.
We first remark that convergence of mappings need not imply convergence
of their complex dilatations. More precisely, let (fe) be a sequence of quasiconformal mappings of a domain A. We suppose that the complex dilatations
4. Ecluami
29
Equatiofl
p, of
satWy
oation
uniformly in A towards
p. By Theorm 22,
k <1 and that
converges locally
mapping! with complex dilatation
k, but otherwise there need be no connection
p. i.e., the local mapping properties off,, and I

between the func
may be quite
p. 186).
The situation changes if the functions p,, converge.
Theorem 4.5. Let (f,,) be a sequence of K-quasiconformal mappings of A

which converges locally uniformly to a quasiconformal mapping I with complex dilatation p. If the complex dilatations
of f,, tend to a limit a.e., then
urn
p(z) a.e.
This result ([LV], p. 187) is needed to take the third step in the proof of
Theorem 4.4 sketched above. It can also be used to prove that an arbitrary
quasiconformal mapping can be approximated by smooth quasiconformal
mappings (cf. [LV], p. 207).
The following complement to Theorem 4.5 shows that convergence of

complex dilatations implies convergence of the corresponding normalized
mappings.
Tbeorein 4.6. Let p and p.,, n = 1, 2, ..., be measurable functions in the plane
such that
k < I and limp,,(z) = p(z) ce. 1ff and f,, are the quasiconformal mappings of the plane which fix the points 0, 1 and and have the
complex dilatations p and p,,, then f(z) = urn f,,(z) uniformly in the plane in the
spherical metric.
PROOF. By Theorems 4.4 and 4.2, the mappings f and f,, exist and are uniquely determined. By Theorem 2.1, (f,,} is a normal family. By Theorems 4.5
and 4.2, every convergent subsequence (4) tends to f. Then the sequence (f,,)
itself has the limit f.
4.7. Decomposition of Quasiconformal Mappings

Let f be a quasiconformal mapping with maximal dilatation K, and assume
that f f2of3, where f1 and f2 are K1- and K2-quasiconformaL We then
have trivially K K1K2. Using Theorem 4.4 we shall now show that for
any given K1 K, a "minimal" decomposition 1=12
always exists with
K = K1K2.
Theorem 4.7. Let f be a quasiconformal mapping with maximal dilatation
K, and 0 < t < L' Then f = 12011, where 11 is
K
PROOF. Let p denote the complex dilatation of f. We choose the complex

dilatation off1 as follows: p1(z) is the point on the line segment from 0 to
1. Quaaiconfonnal Mappinp
30
p(z) for which h(O,p1(z)) th(0,p(z)), where h is the hyperbolic distance in

the unit disc. It then follows from formula (1.1) that
1+
1 fPi(z)I
(1 +
(4 16)
From this we see that 11 is Kt-quasiconformal.

If p2 denotes the complex dilatation of f2 =
then by formula (4.4),
Hence, again by (1.1),
= 2h(p1(z),p(z)) =
2(1 t)h(0,p(z))
(1
t)log
We conclude that f2 is K1'-quasiconformal.

It follows from Theorem 4.7 that 1ff is a K-quasiconformal mapping and
> 0 is given, we can always write
f=fAo" 012011,
where each mapping
(4.17)
i = 1,2,..., n, is (1 + e)-quasiconformal.

5.1. Boundary Function of a Quasiconformal Mapping
A quasiconformal mapping f of a Jordan domain A onto another Jordan
domain B can always be extended to a homeomorphism between the closures
of A and B (ci. 2.1). Thus we can speak of the boundary function off, and it
is a homeomorphism of 8A onto 8B.
Now let h: A -, aa be a given homeomorphism under which positive
orientations of the boundaries with respect to the Jordan domains A and B
correspond to each other. The boundary value problem is to find necessary
and sufficient conditions for Is to be the boundary function of a quasiconformal mappingf: A -+ B.

We restrict ourselves here to studying the normalized case in which A =
B = the upper half-plane, which we denote by H. Then the given mapping h
is a homeomorphism of the one point compactification R of the real axis
onto itself:
Let x1, x2, x3, x4 be a sequence of points of R determining the positi
orientation with respect to H. We call
5. The Boundafy Value Problem
M(H(h(x1), h(x2),h(x3), h(x4)))

SUP
where
M(H(x1,x2,x3,x4))
the supremum is taken over all such sequences x1, x2, x3, x4, the
maximal dilatation of h. If it is finite, h is said to be quasiconformal, and if it is

K, the mapping h is K-quasiconforinal. From the definition it is clear that
these one-dimensional quasiconformal mappings have the customary properties of quasiconformal mappings in the plane: If h is K-quasiconformal,
I = 1, 2, then h2 oh1
then so is its inverse
and if is
is K1 K2-quasiconformal.
Now suppose that h is the boundary function of a K-quasiconformal

mapping f: H H. Then clearly h itself must be K-quasiconformal, and we
have found a necessary condition for 1*, albeit an implicit one.
This necessary condition becomes much more explicit if we choose the

points x1, x2, x3, x4 in a special manner and introduce the normalization
The normalization means that h is a strictly increasing continh(cx) =
uous function on the real axis, growing from
to
Theorem 5.1. The boundary values h of a K-quasiconformal self-mapping f of
the upper haif-plar f(co) =
satisfy the double inequality
1/).(K)
7z(x + t) h(x)
h(x) h(x t)
(5.1)
A(K)
for all x and all t > 0. Here 2 is the distortion function of 2.4. The inequality is
sharp for any given K, x and t.
PROOF. Choose x1 = x
t,
co, and denote the

in 2.4, we then have
x2 = x, x3 = x + t, x4 =
middle term in (5.1) by

By the
M(H(x1,x2,x3,x4)) = ?nd
1
M(H(h(x1),h(x2),h(x3), cx))
for s =
and s =
1.
31.1/2))
Thus (5.1) follows from the fact that h is K-quasi-
conformal. From the characterization of 2 as an extremal function we deduce

that (5.1) is sharp.
0
5.2. Quasisymmetric Functions

An increasing homeomorphism h: ii
with h(cx)
quasisyminetric if
h(x + t) h(x)
h(x)h(xt)
is
said to be k-
I. Quaakon(ornial Mappings
32
for all XE P and all t >0. A function is quasisymmetric if it is k-quasisymmetric for some k. The smallest possible k is called the quasisymmetry constant of h. From the preof of Theorem 5.1 it follows that if h is
K-quasiconformul and h(oo) = oo, then h is A(K)-quasisymmetric.
Lemma 5.1. The family of k-quasisymmetric functions h which keep 0 and 1
fixed is equicontinuous at every point of the real axis.
PROOF. We first conclude from
that
(5.2)
for every non-negative integer n. In looking for a bound for h(a + x) h(a)
we may assume that a and x are non-negative. Then, if x 2", it follows

from (5.2) that
" k
(5.3)
i)
If m
a<
h(a
+ 1, where m is
an integer, then
+ 1) h(a) km(h(a + I
m) h(a m))
kmh(2)
k'(k + 1).
Hence, (5.3) implies equicontinuity at a.
A quasisymmetric function which
fl
fixes 0 and 1
is said
to
be
We conclude that every infinite sequence (he) of normalized k-quasisymmetric functions contains a subsequence which is locally uniformly convergent on the real axis. The limit is also a normalized k-quasisymmetric
function. This result allows the followir.g conclusion.
Lemma 5.2. Let La, b) be a closed interval on the real axis, and e > 0. Then
there is a > 0 such that for a normalized quasisymmetric function 1*,
xe(a,b],
h(x)xj<.s,
whenever h is (1 + b)-quasisyrnmeiric.
PROOF. If the lemma is not true, there is an e > 0 and a sequence of normalized (1 + l/n)-quasisymmetric functions h0, n = 1, 2,..., such that
sup
x( e
a normal family, there is a subsequence which

converges uniformly on [a,b). The limit is 1-quasisymmetric and hence the
identity. This is a contradiction.
(2
33
5.3. Solution of the Boundary Value Problem

By Theorem 5.1, quasisymmetry is a necessary condition for h to be the
boundary function of a quasiconforinal self-mapping of H fixing oo. The
condition is also sufficient (Beurling and Ahifors [1]):
Theorem 5.2. Let h be k-quasisymmetric. Then there exists a
self-mapping of the upper half-plane which has the boundary values h and whose
maximal dilatation is bounded by a number K(k) which depends only on k and
tends to I ask'!.
PROOF. Letf be defined in the closure of H by the formula
1'l
f(x+iy)=fl
Jo
Jo
(h(x+ty)h(xty))dt.
(5.4)
Clearly f = h on the real axis. We calif the BeurlingAhifors extension of h

and prOve that it has the desired properties.
Set
c4x,y)=
h(t)dt,
(5.5)
P(xiY)=J
0
h(t)dt.
Y
and fi are continuously differentiable in H, and an easy calculation shows that the Jacobian J
oj, is positive throughout H. More
than that, we deduce from (5.5) that f is a continuously differentiable bijective
self-mapping of H (cf. [LV], p. 84). This conclusion can be drawn from the
fact that h is a homeomorphism of R onto
quas symmetry is not needed
We see that
here.
In estimating the maximal dilatation of (5.4) we make use of linear functions z -.

= ajz + j = 1, 2, with real coefficients > 0 and 1ff is
the BeurlingAhifors extension of h, then A2ofoA1 is the BeurlingAhifors
extension of A2 oh o A1 1k. This can be verified directly from (5.4). Moreover,
the maximal dilatation of f and the quasisymmetry constant of h do not

change under such a transformation.
Suppose now that there is not a number K(k) bounding, the maximal
dilatation of an extension (5.4). Then there are k-quasisymmetric functions\
and points ; H such that the dilatation quotients I), of the Beurling-.
of have the property
-. (cf. Theorem 3.1).
Application of suitabl linear transformations 4 makes it possible to assume that the boundary'functions hN are normalized and that; =; l for every
n. By Lemma 5.1, the functions h,, then constitute a normal family. We may
Ahifots extensions
34
1.
thus suppose that the functions h converge to a k-quasisymmetnc function
h, uniformly on bounded intervals of the real axis. Let f denote the Beurling
Abhors extension of this limit function h.
=
1),
Let s,,, /1,, be the functions (5.5) for
From
1,
and
ro
(xA),(i) =
h,,(t)dt,
1)
J1
at i converge to the
conclude that the partial derivatives of and
+
partials of and at i. Because J(D + lID) =
+
+
it follows that
tends to the dilatation quotient of f at
+
is impossible, and the existence of a finite bound K(k)
i. Hence, DR(i)
we
follows.
The above reasoning can also be used to proving the existence of a bound
K(k) with the additional property K(k) 1 as k ' 1. If no such K(k) xists,
there is a sequence of normalized functions
whose quasisymmetry constants tend to I while
does not converge to 1. From Lemma 5.2 we
conclude that lim
h(x) = x. By (5.4), the BeurlingAhifors extension
of the limit function is then the identity mapping. Thus urn
D(i) = 1,
and we have arrived at a contradiction.
0
For all our applications it is sufficient just to know the existence of a

bound K(k) which is finite and tends to I as k . 1. For the sake of completewe mention here that K(k) can be estimated. Beurhing and Ahlfors [1]
show, after rather laborious computation, that if the imaginary part of (5.4)
is multiplied by an appropriate positive constant, the maximal dilatation K
of the modified extension satisfies the inequality K k2. For the required
calculations, see also Lehtinen [1).
The smallest upper bound known at present for the maximal dilatation
in the class of quasiconformal mappings with k-quasisymmetric boundary
values is
2k 1)
(Lehtinen [3]).
5.4. Composition of BeurlingAhifors Extensions

and 12 their Beurling
Let h1 and h2 be k-quasisymmetric functions and
Ahifors extensions. If the quasisymmetry constant of h2 o hj' tends to I, it
follows from the proof of Theorem 5.2 that the maximal dilatation of the
BeurhingAJ)Ifors extension of h2 oh;' converges to 1. However, 12 ofr' is
the BeurlingAhifors extension of h2 o hj'. A small modificanot
of Theorem 5.2 is therefore required to show that the
tion
then also tends to 1. (Cf. Earle and Eels [1].)
maxidM
of 12
which we shall need in 111.3.23.
This is
35
5. The Houndary Value Problem
be k-quasisymmetric functions and and 12 their

tends
BeurlingAhifors extensions. If the quasLcymmetry constant of h2
to 1, then the maximal dilatation of 12 off' converges to 1.
Lemma 5.3. Let hL and
h2
PROOF. Suppose the lemma is not true. Making use again of the linear
transformations 4,, appealing to Lemmas 5.1 and 5.2, and reasoning as in the
proof of Theorem 5.2, we would arrive at the following situation. There exist
n
normalized k-quasisymmetric mappings
and
1, 2, ..., which
converge locally uniformly on the real axis to k-quasisymmetric mappings h1
o
converges locally uniformly to the identity
and h2 and for which
and fn2 are the BeurlingAhifors extenmapping. On the other hand, if
sions of
and
the dilatation quotient of 1,2
at the point i is
bounded away from 1.
be the Beur4ing-Ahlfors extensions of h1 and h2. Arguing as
Let f1 and
in Theorem 5.2 we obtain the result
limD1201-1(i) =
it follows that h1 = h2. Hence 11 = 12' and so
But from
D120111(i) = 1.
Thus (5.6) is a contradiction.
' x
(5.6)
5.5. Quasi-Isometry
In a later application, the following property of the mapping (5.4) will be

needed (Ahlfors [4]).
Lemma 5.4. The BeurlingAhifors extension of a quasisymmetric function is a
quasi-isometry in the hyperbolic metric of the upper half-plane.
PROOF.
Let h be a k-quasisymmetnc function and f its BeurlingAhlfors
extension (5.4). We have to prove the existence of a constant c depending only

on k, such that
I
IdzI
clmz
Idf(z)I
IdzI
Imf(z)
Imz
(5.7)
The mapping f is K-quasiconforrnal for a K which depends only on k. We

have
KJ1(z), where J1 is the
and maxj
maxj
Jacobian off. Hence, the ript-hand inequality (5.7) follows if we prove that
y2KJ1(z) c2(Imf(z))2
(5.8)
with y = Im z.
Suppose that (5.8) is not true, and apply again the same method of reasoning used in the proofs of Theorem 5.2 and Lemma 5.3. We conctude that
there are normalized k-quasisymmetric functions ha, which converge locallY
36
1. Quasiconformal Mapping
uniformly to a k-quasisymmetric function h, such that for the Beurling

Abifors extensions
of h, then
-. J1(i), Imfft(i)
Imf(i), as we showed in the proof of Theorem 5.2. It follows that (5.9) is

impossible.
Similar reasoning yields the lower inequality (5.7).
5.6. Smoothness of Solutions

In the 1950's it was a famous open problem whether the boundary function
of a quasiconformal self-mapping of H is absolutely continuous. A direct
construction (BeurlingAhlfors [1)) combined with Theorem 5.2 gives an
entirely negative answer:
For every k> I, there is a k-quasisymmetric function h which is singular, i.e.,
for which h'(x) = 0 a.e.
Hence, a quasiconformal mapping, while absolutely continuotison lines,

need not be absolutely continuous on every closed line segment in its domain
of definition. Singular quasisymmetric functions were first regarded as a

curiosity. But now we know that, except for linear mappings, all boundary
functions encountered in the classical theory of Teichmller spaces are singular (see V.3.6).
In spite of the fact that a given boundary function may be singular, it

always admits extensions which are very smooth in the upper half-plane. We
remarked already that the solution (5.4) is continuously differentiable in H,
and very much more is true:
Theorem 5.3. For every quasisymmetric function, the boundary value problem
has a real-analytic solution.
PROOF. The result can be proved with the aid of the decomposition formula
function
(4.17) in 4.7 which makes it possible to express a
as a composition of (1 + e)-quasisymmetric functions. The proof based on
t proof (Lehtinen [1]) is
this method will be given in 11.5.2. A more
obtained by a modification of the formula (5.4). e can write in (5.4)
Ref(z)
ic(t)(h(x + ty) + h(x ty))dt,
=
on [0, 1] and vanishes elsewhere, and a similar expression
where ,c(t) = I
obtains for im f(z). If K is replaced by a suitable exponential, the corresponding f turns out to be a real-analytic solution.
U
__LI_
S. The Boundary Vakie Problem
37
5.7. Extremal Solutions

Let Fh be the family of all quasiconformal self-mappings of.H whose boundary values agree with a given quasisymmetric h. Then F,, is a countable union
of normal families, and each of these contains its limits under locally uniform
convergence. This allows an important conclusion:
In the
class
F,,, there always exists an extremal mapping which has the
smallest maximal dilatation in F,,.
While the existence of an extremal mapping can be deduced immediately,

it
is much more diflicult to study its uniqueness. It was in fact an open
problem for quite a long time whether the extremal mapping in F,, is always
unique, until
in 1962 gave an example which shows that Fh can
contain more than one extremal.
In Strebel's example one considers the domain A which is the union of the
lower half-plane and the "chimney" {zllmz O} {zlO < Rez < 1} (Fig. 1).
In A we set f1(x + ly) = x + iKy, K> I, and define another mapping 12 so
that 12 agrees with f1 in the chimney and is the identity in the lower halfplane. Then f1 and f2 are K-quasiconformal in A, they agree on the boundary
of A, and both are extremal for their boundary values (Strebel [1]). By using
a suitable conformal mapping of A onto H we can transform f1 and 12 to
normalized self-mappings of H.
More examples of boundary values with non-unique extremals will be
obtained in V.3.7. On the other hand, there are important cases in which
uniqueness can be proved (V.8.5). The question of unique extremality has
been systematically studied by Reich and Strebel; see, for instance, Strebel
[1], Reich and Strebel [1] and Reich [1].
In certain cases the
solution (5.4) is far from extremal.
Lehtinen [3] proved that if h has the quasisymmetry constant k, the maximal
dilatation of (5.4) is always k. Now let h be the restriction to R of the

K-quasiconforrnal extremal mapping described in 2.4. Then h has the quasisymmetry constant
In this case the minimal maximal dilatation in F,, is
equal to K, whereas by (2.6), the maximal dilatation of the BeurlingAhifors
extension is
1/2.
The preceding considerations make it possible to compare the maximal
dilatation of the boundary function h with the extremal maximal dilatation
for F,,.
Figure 1. Strebel's chimney.
38
Lemma 5.5. Let K* be the maximal dilatation of the quasisymmetric function

h and K the minimal maximal dilatation of the quasiconfonnal self-mappings of
1.1 with boundary values h. Then K
, 1
as K* -.+
1.
by the remark in 5.2 prePROOF. The function h is

ceding Lemma 5.1. Here 1(K*) -+ 1 as KS . 1, as we noted at the end of 2.4.
By Theorem 5.2, the maximal dilatation of the BeurlingAhifors extension of
h tends to 1 as K5 . 1. A fortiori, the same is true of the minimal maximal
dilatation'K.
0
The bounds at the end of 5.3 for the maximal dilatation of the quasiconformal extensions of h yield quantitative estimates between K and K. For
instance, we have
K5 K ).(K)312.
(5.10)
Here the lower estimate is trivial, and the upper estimate follows if we use
Lehtinen's bound for the maximal dilatation of the modified extension (5.4).
For our later applications, the qualitative result in Lemma 5.5 is sufficient.
6. Quasidiscs
6.1. Quasicircles
A Jordan curve can be defined as the image of a circle under a homeomorphism of the plane. If the homeomorphism is conformal, then the image is a
circle. Between the topological and proper circles, quasicircies form a class of
curves which will come to frequent use in Chapters II, III and V.
A quasicircie in the extended plane is the image of a circle under a

quasiconformal mapping of the plane. If the mapping can be talcen to be
K-quasiconformal, the image curve is called a K-quasicircle. A domain
bounded by a quasicircie is called a quasidisc.
Let f be a quasiconformal mapping of a domain A and F a compact subset
of A. Then there exists a quasiconformal mapping of the plane whose restriction to F agrees with f ([LV], p. 96). It follows that f maps circles in A onto
quasicircles.
Since a quasiconformal mapping preserves sets of area zero, a quasicircie
has zero area. On the other hand, it is possible that all non-empty subarcs of
a given quasicircie are non-rectifiable; concrete examples are provided in

[LV], p. 104. Gehring and Visl [1] have proved the striking result that,
while the Hausdorif dimension of a quasicircle is always less than 2, it can
take any value A, I A < 2. We remark that quasicircies with Hausdorif
dimension greater than 1 play a role in the modern theory of iteration of
polynomials in the plane.
It follows from what we said in 2.4 that a homeomorphism of the
39
6. Quuidiacs
plane
which is K-quasiconformal in the complement of a straight line is
K-quasiconfonnal everywhere. From the definition of quasicircies we obtain

immediately a generalization, which we shall need later several times.
Lemma 6.1 Let C be a quasicircie and f a homeomorphism of the plane which
is K-quaslconformal in the complement of C. Then f is a K-quasiconformal
mapping of the plane.
PROOF. It follows from the definition of a quasicircie that there is a quasiconformal mapping w of the plane which maps the real axis onto C. Then
fo w is quasiconformal in the plane, as is also f = (fo w) o
Since the area
of C is zero, we conclude from Theorem 3.5 (Analytic definition) that f is
K-quasiconformal.
0
6.2. Quasiconformal Reflections
Let C be a Jordan curve bounding the domains A1 and A2. A sense-reversing
of the plane which maps A1 onto A2 is a
K-quasiconformal involution
K-quasiconformal reflection in C if q, keeps every point of C fixed.
Theorem 6.1. A Jordan curve admits a quasiconformal reflection if and only if
it is a quasicircle.
PRooF. Suppose first that C is a quasicircie. Let f be a quasiconformal map-
ping of the plane which maps A1 onto the upper half-plane H. Then the
mapping =
ojof, wherej(z) = is a quasiconformal reflection in C.
Conversely, let be a quasiconformal reflection in a Jordan curve C. Let
h map H conformally onto A1. Define f
= h(z) in the closure of H,
f(z) =
in the lower half-plane. Then f is a homeomorphism of the
plane which is quasiconformal off the real axis, which it maps onto C. By
Lemma 6.1, f is quasiconformal in the plane, and so C is a quasicircle.
We can draw certain additional conclusions from the above proof.

First, if C admits a K-quasiconformal reflection, then C is a K-quasicircle.
In the opposite direction we deduce that a K-quasicircle always admits a
K2-quasiconformal reflection.
In the second part of the proof, the required quasiconformal mapping of
the plane is conformal in a half-plane. For the sake of later reference, we wish
to express certain connections between conformal mappings and quasicircles
explicitly.
Lemma 6.2. A K-quasidisc A has the following properties:
10
Every quasiconformal reflection in OA is of the form fojof1 , where
quasiconfomal mapping of the plane which maps the upper half-plane H

conformally 'onto A, and) denotes the reflection z
40
2 A is the image of H under a K2-quasiconformal mapping f of the
plane
in H.
which
3 Every conformal mapping f: H -' A has a K2-quasiconformal extension to

the plane.
be an arbitrary quasiconformal reflection in 8A. 1ff is constructed as in the second part of the proof of Theorem 6.1, with A replacing
A1, we obtain 1. Since OA is a K-quasicirck, there exists a K2-quasiconformal reflection q'. The corresponding f is also K2-quasiconformal, and 20
follows. We use this same in 3 and conclude that ofoJ is a desired
extension of f.
0
PROOF. Let
We return'to our previous notation and denote by C a Jordan curve which

bounds the domains A1 and A2. In what follows, c(K), c1(K), ... denote
constants which depend only on K.
Lemma 6.3. Let
be
a K-quasiconformal reflection in C which passes through
x.Then
c1(K)
(6.1)
'12(4)(Z))
where
is
the Poincar density of A2.
PRooF. Let h: H i A1 be a conformal mapping, h(ao)

and set again
f = h in the closure of H and f = q, oh of in the lower half-plane. Then f is a
K-quasiconformal mapping of the plane.

Fix zeA1 and z0eC. The function f maps the circle {wJIw
onto a curve which passes through the points z and
By Theorem 2.4,
Iz ,zoVc(K)
If we choose
such
frp(z)
that
z01
zol
c(K)Iz
zol.
(6.2)
is equal to the distance d((p(z), C), then
+ Iz
z01 (1 + c(K))d(o(z),C).
From this we obtain (6.!) by using the inequality (1.3).
We remark that (6.2) yields the estimate
d(z, C)/c(K) d(p(z), C) c(K)d(z, C).
(6.3)
Here we can take c(K) = A(K) (see 2.5).
For our later applications (see 11.4.2) it is important to know that there
exist quasiconformal reflections whih are quasi-isometries in the euclidean
metric (Ahlfors [41). We also call such reflections Lipschitz-continuous.
and A2 and
Theii there exists a c2(K)-quasiconformal reflection q' in C,
Lemma 6.4. Let C be a K-quasicircle bounding the domains

passing through
41
tQuasidlacs
continuously differentiable in A1 and A2, such that
Idq(z)J c3(K)Jdzl
(6.4)
at every point zeA1.
PROOF. Let h1 be a conformal mapping of A1 onto the upper half-plane H

of A2 onto the lower half-plane, both fixing oo.
and h2 a conformal
extensions to the boundary, we can
Since h1 and h3
on the real axis. It is quasisymmetric. For if is a
form the
reflection in C and j again denotes the reflection z ' 2,
then Jo h2 o o
a quasiconformal self-mapping of H with boundary
values
By using the BeurlingAhlfors extension (5.4), we construct a quasi..
conformal diffeomorphism f: H -+ H with boundary values h2 o hj'. Set
in A2. Then q' is
ojofoh1 in the closure of A1 and q =
=
=
a c2(K)-quasiconformal reflection in C, which is continuously differentiable
outside C.
Since the Poincar metric is conformally invariant, it follows from formula
(5.7) that
172(q)(z))Id(p(z)j
From this we obtain (6.4), in view of(l.3), (1.4) and (6.3).
6.3. Uniform Domains

Let us now start studying the geometry of quasidiscs. We shall prove a chain
of theorems which, when put together, give several characterizations for

quasidiscs and shed light on their geometric properties. In subsections 6.3
6.5, we follow the presentation of Gehring [4). A summary will be given at
the end of subsection 6.7.
Throughout this subsection we assume that A is a simply connected proper
subdomain of the complex plane. The domain A is said to be uniform if there
are constants a and b such that each pair of points z1, z2 e A can be joined by
an arc ri in A with the following properties:
1 The euclidean length of satisfies the inequality
l(x) aizi
(6.5)
2 For every z e
min(I(x1), 1(a2)) bd(z, A),
of a\{z).
where
It will appd
first prove directly
(6.6)
if aild only
is a quasidisc. We
is tsnifdrm. This requires a fairly lengthy
L Quasiconformal Mappiap
42
argument, and we begin with a lemma in which, as before, c(K), c1(K),

constants depending only on K. These are not necessarily the same
that arose in the previous subsection.
Lsra6S.Let AbeaK-quasidisc with

=
mappMg
Then
1"
aconformai
(K)d(f(iy), 8A)
If'QOl di
Jo-
forO<y <
By Lemma 6.21 has a K'-quasiconformal extension to the plane. We
may assume without loss of generality that f(O) = 0.
For y >0, the mappingf satisfies the inequality
.PKOOF.
If'(iy)l
4d(f(iy),
(6.7)
This follows directly from the Koebe one-quarter theorem cited in 1.1, if we
apply it to the function C . (f(iy + yC) f(ay))/yf'(iy).
In order to estimate the distance d(f(iy), .3A) we fix y. After this, we

y and that IfQy,)I =
choose a sequence (yj) so that 0 <
c = c(K2) is a constant for which Theorem 2.4 holds forf.
(We can take, for instance, c =
Because d(f(it),
t
Let Yj+1
d(f(it), OA) clf(iyj)
If(it)
1(0)1
1(0)1,
we obtain from (2.8)
If(iy)f.
Hence, by (6.7)
I
If'(it)Idt
Yj+i
The Iogarithpi can be estimated by aid of (2.9). Let n be the smallest mteger
1(0)1, and so by (2.9),
for which c n. Then If('y,) f(0)f
01
01
c2(K)yj1.
It follows that
I
lf'Qt)idt
Jo
Finally, if xe oH, then by applying (2.8) again we infer that If(iy)I

clf(ly)
Thus lf(iy)I cd(f(iy),OA), and the proof is completed.
Let z1 and be points of A and the hyperbolic segment of A joining z1

and z2. If A is a disc or a half-plane, it is easy to show that (6.5) and (64) hold
n/2. This property generalizes to quasidiscs.
for a
43
6. Quaaidiscs
Theorem 6.2. Let A be a K-quasidisc and a a hyperbolic segment In A with

endpoints and z2. Then a satisfies the conditions (6.5) and (6.6) wIth constants
a and b which depend only on K.
PROOF. By Lemma 6.2, there is a K2-quasiconformal mappingf of the plane
which maps A conformally onto the unit disc D. We can choose f so that
and f(z2) are real. Let D' be the open disc in D which has the line segment with endpoints 1(z1) and 1(z2) as a diameter. Then f1(IY) is a bounded
K2-quasidisc, in which a is a hyperbolic line. Since d(z,f'(OD'))

we may assume, therefore, that A itself is bounded and a is a
for
hyperbolic line in A, i.e., z1, z2 EM.
Let A' and a' be the images of A and a under the Mbius transformation
z .-+g(z) = (z
22). Then
1(a)
2)
z2)(w
1)_2,
(6.8)
lwi 12
In order to estimate the

we use the arc length representation
for a'. Let s0 = c1(K)/(c1(K) + 1), where c1(K) is the constant of
Lemma 6.5. IfO <s s0, then lw(s)
1 Iw(s)l 1 s i/(c1(K) + 1).
we apply Lemma 6.5 to a conformal mapping f: H -. A' fixing
For s >
Then a' is the image of the positive imaginary axis, and so by
0 and
s , w(s)
Lemma 6.5,
c1(K)d(w(s),DA')
Jo
for iy = f'(w(s)). From 1 =

lw(s) 11. It follows that
we further conclude that d(w($,t3A')
+ l)2ds + J C1(K) ds = 2c1(K)(c1(K) + 1).

Thus (6.5) is obtained from (6.8) with a = 2c1 (K)(c1 (K) + 1).
In order to establish (6.6), we consider a K2-quasiconformal mapping f of
the plane, f(oo) = oo, which maps A conformally onto the unit disc D. Fix.
za and choose z0e.3A so that 12
= d(z,c3A). Since f(a) is a hyperbolic
line in D,
mm
Jf(z)
Iz
zjl
2d(f(z), aD) 2 11(z)
By formula (2.9),
mm
J=1.2
2c(K2)21z
Zol
2c(K2)2d(z,A).
0,
Since I(aj) alzzjl by(6.5), we obtain (6.6) with b = 4c(K2)2c1(K)(c1(K)+ 1).
The following result follows immediately from Theorem 6.2.
Theorem 6.3. A quasidisc is a uniform domain.

Theorem 6.3 is the first link in a closed chain of four theorems. 013cc these
have all been proved we get the converse to Theorem 6.3.
6.4. Linear Local Connectivity

A set E is linearly locally connected if there is a constant c such that the
following two conditions hold for every finite z and every r> 0, where
D(z,r) {wJIwzI <r}:
1 Any two points of the set En D(z,r) can be joined by an arc in En D(z,cr).
Any two points of the set E\D(z, r) can be joined by an arc in E\D(z, r/c).
20
In order to illustrate this notion, we let E be the parallel strip {x +

<y < I). The points 0 and 2r > 0 of E\D(r, r) can be joined in E\D(r, nc)
only if c> r. Letting r ,
we conclude that is not linearly locally
connected.
I
If a simply connected domain A with more than one boundary point is

linearly locally connected, then A is a Jordan domain. To prove this, we
consider a finite boundary point z of A. Let U be an arbitrary neighborhood
of z, and choose r > 0 such that the closure of the disc D(z, Cr) lies in U. Then
V = D(z, r) is a neighborhood of z such that A n V lies in a component of
A
U. It follows that A is locally connected at z. A similar reasoning, based
on the use of condition 2, shows that if e
then A is locally connected
at
We conclude that A is a Jordan domain (see 1.2 or Newman [11
pp. 167 and 161).
We shall see after completing our chain of theorems that in fact, a simply
connected domain with more than one boundary point is linearly locally
connected if and only if it is a quasidisc. We shall now establish'the second
link of the chain.
Theorem 6.4. A uniform domain A is linearly locally connected.
PROOF.
Fix a finite z0 and r > 0, and suppose that
Since A is uniform, there exists an arc joining z1 and

I(ct) oIz1 z21 2ar. IfzEx, we thus have
(zz0{jzz11+1z1
z2 A n D(z0, r).
in A such that
1)r.
It follows that
and z2 in A n D(z0,cr) if c = 2a + 1.
Next assume that z1, z2 e A\D(z0, r). We consider an arc joining z1 and
z2 in A with
for every
Set c = 2b + 1. If joins z1
and z2 in A\D(z0,r/c), the theorem is proved. Now suppose that does not
join z3 and z2 in A\D(z0,r/c). We prove that, nonetheless, A\D(z0,r/c) is

connected.
45
6. Quasidiscs
By our hypothesis, there is a point z
for which z
z0
<nc. For
j = 1, 2 we have
Izj
zf
Izj
z0(
r nc.
Thus
We conclude that D(z0,r/c) c A. It follows that

in A\D(z0,r/c).
and z2 can be connected
From the proof we see that A is linearly locally connected with the constant c 2 znax(a, b) + 1.
6.5. Arc
Condition
Let C be a Jordan curve and z1, z2 finite points of C. They divide C into two
arcs, and we consider the one with the smaller euclidean diameter. The curve
C is said to satisfy the arc condition if the ratio of this diameter to the distance
is bounded by a fixed number k for all finite z1, z2C.
fz1 -A circle satisfies the arc condition for the constant k = 1. An example in
> 0,
the opposite direction is obtained if we consider the curve C = {x +
z2f = 2x2, and the smalx2}. liz1 = x + ix2, z2 x ix2, then
y
ler diameter in the above definition is x. Hence d/1z1 z21 .
as x ' 0,
so that this curve does not satisfy the arc condition.
We shall now establish the third link of our chain.
Theorem 6.5. Let A be a simply connected domain whose boundary contains

more than one point. If A is linearly locally connected, then A Is a Jordan
curve which satisfiec the arc condition.
PRooF. In 6.4, after defining the notion of linear local connectivity, we proved
that
s a Jordan curve.
Choose two finite points z1, z2eA and set z0 = (z1 + 22)/2, r =
1z1
The theorem follows if we prove that at least one of the arcs
a1, a2 into which the points z1, z2 divide lies in the closure of the disc
D(z0, c2r).
The proof is indirect. Suppose there is a t> r and points E ct1\D(z0, c2 t),
I = 1,2.
< <S2 <i; then and z2 belong to the set
A is a Jordan domain, its boundary points are accessible. We can thus
find points z eA D(z0, and arcs
z to z1 in A ri D(z0, s1). By the
linear local connectivity of A, the points
and 4
can be joined by an arc fl3
in A D(z0,cs2).
The points w1 and w2 lie in A\D(z0,c2s2). Therefore, we can find an arc y
46
joining w1 and w2 in A\D(z0,cs2). But then the cross-cut y does not meet the
cross-cut u fl2 u fl3. This is a contradiction, because their endpoints are in
the order
Wi, 22, w2 on A.
The proof indicates that if A is linearly locally connected with a constant

c, then A satisfies the arc condition with the constant c2.
6.6. Conjugate Quadrilaterals

We shall now close our chain of theorems by proving that if the boundary of
a Jordan domain A satisfies the arc condition, then A is a quasidisc. This is a
difficult step for which no simple proof seems to exist. As a preparatory

result, we give a characterization of quasicircies in terms of quadrilaterals.
Let C be a Jordan curve bounding the domains A1 and A2. Take a sequence of four points z1, z2, z3, Z4EC
that A1(z1,z2,z3,z4) is a qua-
drilateral. Then A2(z4,z3,z2,z1) is also a quadrilateral, and these two

quadrilaterals are said to be conjugate.
Lesmna 6.6. Let C be a Jordan curve such that for all conjugate quadrilaterals
A1, '42 with M(A1) = I we have M'.42) K. Then C is a c(K)-quasicircle,
where c(K) depends only on K.
PRooF. Let gi: A1
-, If and 92: '42p H' be conformal mappings, where H' is
the lower half-plane. Consider the increasing homeomorphism x h(x) =

92(9j1(x)) of the real axis. For all quadrilaterals H(z1,z2,.z3,z4) with module
iwe then have

If I( M(H(h(z1),h(z2),h(z3),h(z4))) K.
We proved in section 5 that the validity of this module inequality is

'S sufficient condition for the existence of a c(K)-quasiconformal mapping
[H-.H with boundary values h. Then fog1 extended by 92 iS a c(K)
quasiconformal mapping of the plane carrying A1 onto H. Thus C is a
c(K)-quasicircle.
In order to utilize Lemma 6.6, we need a result about the geometry of

conformal squares.
Lenma 6.7. Let Q(z1, 22,23, z4) be a quadrilateral with module 1, and let
and
denote the euclidean distances in Q between the sides (ii, z2), (tp,X4) and
(z4,
respectively. Then
,.
>
We
z1) in Q
assume that among the arcs which join the sides (22,Z3) and
is a of length
Let 1 be the point which divides Vo into
47
6.
two parts of length 52/2. Set pfr) = 2/s2 if Iz z01 52/2, p(z) = 1hz zoj if
s1 + s2/2, and p(z) 0 elsewhere. The area
s2/2 < Iz
of Q in
this p-metric then satisfies the inequality
it(l 2log(1 + 2s1/s2)).
Consider next an arc y joining the sides (z1,z2) and (z3,z4) in Q. FQr the
p-length of y we obtain a minorant if we integrate 1/x over a segment with
endpoints s2/2 and s2/2 +
Therefore,
log(1 + 2s1/s2).
p(z)IdzJ
-'1
Setting
F(x)
+ 2 log(1 + x)
(log(1 + x))2
we thus we have by formula (1.7),
.1 = M(Q(z1;z2,z3,z4))
From this we obtain, by interchanging the roles of s1 and
2
6.7.
>
Characterizations of Quasidiscs
We can now establish the remaining link of our chain.

Theorem 6.6. A Jordan domain whose boundary satisfies the arc condition is a
quasidlsc.
PROOF. Let C be a Jordan curve which satisfies the arc condition and bounds
the domains A1 and A2. Choose four points z1, z2, z3, z4 on C such that.
with module 1. We shall derive an upper
bound for
quadrilateral A3(z4,z3,z2,z1).
Let
tl)c distance in A1 between the sides (z1,z2) and (z3,z4), and
d1 the same distanee measured in the plane. For the remaining sides (z2,z3)
and
these distances are denoted by and d2. From Lemma 6.7 it
follows that
s1 > 103d2.
(6.9)
Since C satisfies the arc condition, there is a constant k I such that one
of the sides (z2,z3), (z4,z1) lies inside a disc of diameter ku1. From this we
conclude that
103,rk
(6.10)
48
For if not, one of the sides (z2,z3), (z4, z1) lies in a disc of diameter 103d2/*.
The other, which is at a distance d2 from this one, must lie.outside this disc.
can be joined in A1 by a circular
It follows that the sides (z1, z2) and (z3,
This is in contradiction with (6.9), and (6.10) follows.
arc of length
Since (6.10) is formulated in terms of the distances in the plane, it can be
used to estimating the module of A2. We first conclude as above theexistence
of a disc
<kd2f2 which contains one of the sides(z1, z2), (23, 24). Let
r=
kd2/2
the
and define p(z) = I if tz z01 <r, and p(z) = 0 elsewhere. By
103d2/irk.
p-length of an arc yjoining the first and the third side of A2 is
Hence,
IO6ir(irk2/2 + 10-3)2.
M(A2(z4,z3,z2,z1))
By Lemma 6.6, C is
a quasicircie, and the theorem is proved. We remark that
C is a c(k)-quasicircle where c(k) depends only on the constant k in the arc

condition.
0
Before summarizing our results we remark that the converse of Theorem
6.6, which we now know to be true, admits a fairly simple direct proof ([LV],
p. 101).
satisfy a particularly
Quasicircies passing through
simple geometric
condition.
Theorem 6.7. Let C be a K-quasicircle passing through

points of C such that z2 Ues between z1 and z3. Then
and zj, Z2, 23 finite
Izi
PROOF. Let f be
c(K)Izj
+ 1z2
(6.11)
a K-quasiconformal mapping of the plane which maps the
i = 1, 2, 3, and
axis onto C such that f(co) =
Denote x1 =
= lxi x21}, C2 = {wflw _X3j = fx2 x31}. Join z1 and
byatit segment L, and denote by a1 and a2 the first points at which L
when one moves along L from z and from z3. Then
f(C
real
jE1
a1.
1z2
z31 c(K)1z3 a2l.
These yield the desired inequality (6.11). It follows from the proof and our
remark in 2.5 that (6.11) holds for c(K) = .a(K).
0
CnvlsCly,
a JOfdan CUrve C containing oo satisfies (6.11).
Then C satisfies the arc condition with the constant c(K), and C is a
The implications in
form an unbroken chain which
I
6, Quasidiscs
49
starts and ends with quasidiscs. This makes it possible to draw the remarkable conclusion that the converse of each of these theorems is true. It follows
that quasidises can be characterized as domains with the hyperbolic segment
property of Theorem 6.2, as uniform domains (Theorem 6.3), as domains

which are linearly
connected (Theorem 6.4) or as domains whose
boundary satisfies the arc condition (Theorem 6.5 or Theorem 6.6). Here we
assume of course that the domains are simply connected and have more than
one boundary point. In addition, Theorem 6.1 tells that quasidiscs are Jordan
domains whose boundaries admit quasiconformal reflection.
The special position of infinity necessitates some clarification. In defining
uniform domains A we assumed that
is not in .4, and Theorem 6.2 is not
true if A contains
By contrast, Theorems 6.5 and 6.6 hold without this
restriction, which is not utilized in the proofs. It is not hard prove that the
converses of these theorems are also valid for domains in the extended plane
For Theorem 6.6 this follows immediately from the fact that if A is a quasidisc and cc. A, then
is a bounded quasicircie. Since the converse of
Theorem 6.6 holds for quasidiscs of the complex plane,
satisfies the arc
condition.
There are aiso characterizations cf quasidiscs in terms of analysis rather
than geomel ry. They sometimes reveal surprising Connections between var-
ious problems of analysis which on the surface have nothing to do with

quasicouformal mappings. Iii 11.4 we shall establish a result of this type, using
Schwarzian derivatives (see especially 11.4.4). A comprehensive account of the
main properties of quasidiscs known in 1982 is given in Gehring's lecture

notes [4}, which we have utilized to a large extent in this section. These notes
also contain an extensive bibliography.
Quasicircies
mtro'duced by Pliuger
(1961) and Tienari (1962). In
1963, Ahllors
characterized quasicircies geometrically by proving that the
arc condition is necessary and sufficient. In this same paper, he also introduced
reflectionc and used them to prove an important ex-
tension theorem for conformal mappings (Theorem 11.4.1). Gehring [2J

defined the notion of lincaz local connectivity and proved Theorem 6.5 in
1977. Theorem 63 and its cor.verse were established by Martio and Sarvas
[1] (1979). and Thc'rem 6.2 with its converse by Gehring and Osgood [1]
(1979).
CHAPTER II
Univalent Functions
Introduction to Chapter II
The theory of univalent analytic functions covers a large part of complex
analysis. In this chapter, we deal with certain aspects of the theory which are
directly or indirectly connected with Tcichmller theory. The interaction

between univalent functions and Teichnillcr spaces was already explained
in the.tntroduction to this monograph. A more comprehensive description is provided by Chapters II, III, and V, taken together.
In this chapter a central position is occupied by the Schwarzian derivative
of a locally injective meromorphic function. In section 1 we begin with the
classical result that the Schwarzian derivative vanishes identically if and only
if the function is a Mbius transformation. Sonic other basic properties of the
Schwarzian and its
sup-norm are also established.
In section 2 we consider Schwarzian derivatives of conformal mappings of
a simply connected domain A. Particular attention is paid to the case in

which the image domain is a disc. The norm of the Scbwarzian is then
intimately related to the geometry of A and provides a measure of the distance of A from a disc.
Sections 3 and 4 deal with a problem which, apart from its intrinsic interest,
plays an important part in Tcichmller theory. Let f be quasiconformal in

the plane with complex dilatation p and confocmal in a simply connected
domain A. The problem is to describe the inteirelations existing between p
and the Schwarzian derivative of the restrictionf I A.
a quantitative estimate which shows that if the sup
In section 3 we
norm of p is small, then the norm of the Schwarzian of f IA is also small.

The method of proof can be adapted to the study of many other problems
concerning univalent functions with quasiconformal extensions. We establish
51
1. Scbwarzian Derivative
some results of this type, even though they might veer a little off the main
track from the standpoint of applications to the Teichmller theory.
We return to the main problem in section 4. A basic result says that if a
function f, meromorphic in a quasidisc, has a small Schwarzian derivative,
then I is univalent and has a quasiconformal extension to the plane with a
It is then proved that the result does not hold in simply connected
small
domains which are not quasidiscs. In this section, the interplay the concepts complex dilatation, quasidisc and Schwarzian derivative becomes very
concrete.
In section 5 we consider meromoi'pic functions in a disc. Results of

section 4 can then be supplemented and expressed in a more explicit form.
Again, this is not only of interest in itself but leads to important conclusions
in Teichmller theory.
1. Schwarzian Derivative
1.1. Definition and Transformation Rules.
Let us consider a Mbius transformation z -. f(z) = (az + b)/(cz + d). Differentiation yields
f"(z)
f'(z)
(f"'\' (z)
2c
cz + d'
2c2
(cz + d)2
Using the notation
/ ,,\s
conclude that every Mbius transformation satisfies the differential

equation
we
S, =0.
(1.2)
Conversely, if we start from the equation (1.2) and set y = "/f', then
= p2/2. From this we deduce, by an easy integration, that every solution of
(1.2) is a Mbius transformation.
The expression (1.1) is called the
derivative of the functionf. It

can of course be defined for a much more general class of functions than
Mbius transformations. In order to make clear the notions to be used, we
remark that in our terminology, a meromorphic function need not necessarily have poles. A holomorphic (unction is a meromorphic function without
poles. The word "analytic" is a synonym for "meromorphic", but to avoid
confusion it is more rarely used. The terms "conformal mapping", "injective
52
11.
Univalent Functions
meromorphic function", and "univalent analytic (meromorphic) function" all

have the same meaning. Sometimes we follow time-honored practice and say
"univalent" instead of "univalent analytic".
Let us first assume that f is holomorphic in a domain A in the complex

plane andf'(z) 0 in A. We then define the Schwarzian derivative S1 off by
means of formula (1.1).
If, in addition, f(z)
0, we see from (1.1) that

S1(z)
S111(z).
We use this formula to define S1(z) for a meromorphic f at points where! has
a first order pole. It follows that if a meromorphic f is locally injective in A,
then Sf is defined everywhere in A, and it is
in A.
Direct computation gives the transformation rule
= (S1og)g'2
If g is a Mbius transformation, we have 59 =
(1.3)
0,
S109=(Sfog)g'2.
and so
(1.4)
This formula can be used to define the Schwarzian derivative at infinity.

Assume that f is mcromorphic and locally injective in a domain which
contains
and let p be defined in a neighborhood of the origin by
q(z) = I/z). By (1.4), we have z4SQ(z) = S1(l/z). Hence, if we define
= Jim z4S,(z),
z-0
then
is
holomorphie at x. We see that S1 has a zero of order 4 at
infinity. In conclusion, the Schwarzian derivative can be defined in any domain A for every function f meromorphic and locally injective in A, and it is
a holomorphic function in A.
The Schwarzian derivative will play a very central part throughout this
chapter on univalent functions and in the presentation of Teichmiler theory
in Chapters III and V. It was introduced to complex analysis in 1869 by
II. A. Schwarz ([1], p. 78). Schwarz established equation (1.2) and used it to
map a simply connected domain bounded by fmitely many circular arcs conformally onto a disc.
The special role of Mbius transformations in connection with Schwarzian
derivatives appears not only from equation (1.2) but also from the invariance
property following from (1.3) and (1.2): 1ff is a Mbius transformation,
Slog ='S9.
In studying the local approximation of a meromorphic function by Mbius

transformations Martio and Sarvas arrived at the Schwarzian derivative in
a way which justifies using the word "derivative" for the operator
Let f be
a locally injective meromorphic function in a domain A and z0 an arbitrary
finite point of A. Then there is a unique Mbius
h such that
53
1. Schwarvan Derivative
(z 2()
finite. This limit is qual to S1(z0)/6.

for if h corresponds to
In the proof we may assume that
corresponds
to
f.
For
an
undetermined
Mbius
1/f, then z , h(1/z)
z0) + a2(z zo)2 +
transformation h, we write (hof)(z) = a0 +
is
a3(z z0)3
+ ".Since h depends on three complex parameters, there is a
unique h such that a0 = 20, a1 = 1, a2 = 0. Then ((hof)(z) z)/(z z0)3 has

the finite limit a3 as z z0. A direct computation, based on a1 = 1, a2 = 0,
and Sh =0, shows that 6a3 = S1(z0).
1.2. Existence and Uniqueness

The Schwarzian derivative can be prescribed:
Theorem 1.1. Let be a holomorphic function in a simply connected domain A
in the complex plane. Then there is a meromorphic functionf in A such that
S1o.
(1.5)
The solution is unique up to an arbitrary Mbius transformation.
PROOF. Through the stibstitution y f"/f' equation (1.5) transforms to the

Riccati equation y' y2/2 = q'. From this we obtain, by the standard substitution y 2w'/w, the linear second order equation
(1.6)
It is a well-known result in the classical theory of linear differential equa-
tions that given a point z0eA, equation (1.6) has a unique holomorphic
solution w in a neighborhood of z0, once we prescribe the values w(z0) and
w'(z0). It is also easy to verify this directly, with the aid of power series. In
fact, we have q,(z) =
Zn)" in a disc around z0. If
w(z) =
c,,(z
(1.7)
then for w to be a solution we obtain from (1.6)

"2
n(n
0,
n=
2, 3
The coefficients c0 and c1 can be chosen arbitrarily; we take c0
Next we fix an r, 0 < r <

n
0, 1,
1,
md a finite number M> I such that
..., and that Mr2 < 1. Then

s2
n(n
MEr
k0
,t2
ICkI <r'"
k0
rckl.
0, C1 = 1.
<Mr",
54
1, gives
II. Univalent Functions
It follows that (1.7) is a holomorphic solution of (1.6) in
the disc lz zol

Since A is simply connected, we can apply the Monodromy theorem and
obtain a global solution of(1.6) by analytic continuation.
Let w1 and w3 be two linearly independent holomorphic solutions of(1.6).
Since w1w3'
= 0, we have
= constant. Tha constant is
not zero, because w1 and w3 are linearly independent. Set I = w1/w2; then
is a locally injective meromorphic function in A. Direct computation yields
2w/w2 = .
f/f' =
From the invariance of the Schwarzian derivative under Mbius transformations we conclude that, if f is a solution of(1.5) and h an arbitrary Mbius
transformation, then h of also is a solution of (1.5).
Assume, conversely, that I and g are solutions of (1.5) in A. Since f is

locally injective, we can define gof1 locally. Using (1.3) we deduce from
S, = S, that S,01.-1 =0. It follows that locally g = hof, where h is a Mbius
transformation. But then g = h of with the same Mbius transformation h
everywhere in A, and the uniqueness part of the theorem is proved.
We supposed in Theorem 1.1 that

is not in A. If eA, it follows from
the definition of the Schwarzian derivative at
(see 1.1) that Theorem 1.1
remains valid under the sole restriction that the given function p must have
a zero of order 4 at infinity. In particular, a function is determined by its
Schwarzian derivative up to a Mbius transformation.
1.3. Norm of the Schwarzian Derivative

The Schwarzian derivative S1 measures the deviation of I from a Mbius
transformation. In order to make this statement more precise we introduce a
norm for S1.
Let A be a simply connected domain conformally equivalent to a disc and
tjthe Poincar density of A (cf. 11.1). For functions q, holomorphic in A we

define the norm
119'
hA = sup
z.4
In particular,
II
(1.8)
S1 H4
z6A
There are many reasons to use this "hyperbolic sup-norm" instead of the
ordinary norm suplS1(z)l. We shall soon see that (1.8) exhibits more invariance with respect to Mbius transformations than sup(S1(z)(. This is
important as such and becomes crucial when we generalize the notion of
Scbwarzian derivative to Riemann surfaces,because
is a function on
a Riemann surface whereas is1i is usually not.
55
I. Schwazzian Derivative
is important to see how the norm (1.8) transforms under conformal mappings. Let I and g be meromorphic functions in a domain A and h: B -' A a
conformal mapping. By the transformation rule (1.3),
It
(S1o1i
S,oh)h'2.
We also know that the hyperbolic metric is conformally invariant:

oh)
h'
If w = h(z), we thus obtain the basic invariance formula

IS,(w) S,(w)I
(1 9)
?JA(W)
Equation (1.9) yields a number of results about the norm. First, it follows
immediately from the. definition of the norm that we have the invariance
115, S,t =
IIS,Qh S901113.
For the special case in which g = h' is a conformal mapping of A we obtain

the formula
uS,
(1.10)
which will be repeatedly used later. If we choose f here to be the identity

mapping, we get the invariance
IISIIA =
(1.11)
between a conformal mapping and its inverse. Finally, if g h' is a Mbius

transformation, (1.9) shows the invariance of
under Mbius transformations, and (1.10) assumes the form
II
S, HA = uS,0,-1
(1.12)
We see that H S, H is completely invariant with respect to Mbius transforma-
tins: If h and g are Mbius tEansformations, the norms of the Schwarzians

are the same forf in A and hofog in
1.4. Convergence of Schwarzian Derivatives

n = I, 2, ..., are meromorphic and locally
injective in a domain A. If they converge to a locally injective meromorphic
function f locally uniformly in A, then the Schwarzians S1, also tend to S1
locally uniformly in A. However, this does not imply convqgence in norm:
From
Suppose that the functions
liinS,(z) = S1(z)
locally unjformly in A, it does not necessarily follow that
limIIS,S,1L4=0.
(1.13)
11. Univalent Functions
56
A counterexample is obtained as follows. Let r1, 0< < 1, = 1, 2,.

= z + r,,/z,
be numbers tending to I and
{zIIzI> 1). If we act
1(z) = z + lfz, then j, and f are meromorphic and even globally mjective
in A. From IfA(z) f(z)I < 1 r,, we see that fA(Z) .1(z) uniformly even in
the whole'domain A. In spite of this, it follows-from
= (1z12 1)' and
6r,
S1(z)
S1(z)
(z2
that
/
xa
for
(z2
))=6
every n.
The approximating functions
map A onto the exteriors of ellipses which
collapse to a slit domain f(A). However, if a slit domain is approximated

differently, we do get for the Schwarzian derivatives both locally uniform
convergence and convergence in norm.
This is seen from the example in which A is the upper half-plane and
z', where the numbers
are positive and tend increasingly to 2.
Then
and the limit function z f(z) = z2 are univalent in A. From
S,(z) = (1
it follows that
We see that
S,(z)J
4a2
= 21121
S1(z) locally
ifS1
= 2(4
uniformly in A and Xhat
at).
, S1
in norm. In
this case f(A) is the plane slit along the non-negative real axis, and the
approximating domains
are infinite sectors.
= (wIO < argw <
Figure 2 illustrates the difference between the two cases. In order to get the
same limit domain and the same "critical" points, we have replaced in the
first example by
with
+ and in the second one by
+
=
These changes leave the Schwarzians invariant.
The converse problem is to study whether we can conclude from (1.13) that
the mappings converge to 1. Since the Schwarzian derivative determines
Example
1.
S1
Example 2.
S1
Figure
S1 -+
S1
__________
__________
57
b Schwarzian Derivative
-)
z-Plogz
Figure 3
the function only up to a Mbius transformation, Some normalization is of

course needed for the functions and f.
if no normalization is imposed on the functions fM in (1.13), they may
converge pointwise but the limit can be very different from f. For instance,
=
in the upper halfconsider the mappings z ' f(z) = logz and z
Since S1(z) = 1/(2z2), we have
IIS.,, S1 U
2/n2,
does not tend to log z but to the constant 1.

The situation changes if suitable Mbius transformations are applied to f,,
and f. In fact, set
and so (1.13) is valid. But
,r
I+
4n
1we
h(w)
(For the images of the upper half-plane under h0

and h of, see Fig. 3.)
We stiLl have of course Shof '
but now we also have lim
=
h(log z).
In general, the reasoning might go as follows, if we can infer that the

functions f,, constitute a normal family, then there is a subsequence
which
tends locally uniformly to a limit function Jo. If 10 is a locally univalent

meromorphic function, then S1(z) +
for every z e A. On the other
hand, (1.13) implies that
S1(z) in A. Hence, by the uniqueness part of
Theorem 1.I,f = hof0, where his a Mbius transformation.

Let us give an application. Assume that in (1.13) the functions I,, are
conformal mappings of A which have K-quasiconformal extensions to the
plane and which keep three fixed boundary points a1, a2, a3 of A invariant.
By Theorem 1.2.1, the mappings form a normal family. Let be the limit
H. Univalent Functions
58
of a locally uniformly convergent subsequence fm,. In A the function f0 Is

conformal and S,0 = Jim
Hence, we conclUde as above that I = h ofo,
where h is a Mbius transformation. It foUows that f also is a conformal
mapping of A with a quasiconformal extension. If I fixes Oi, 02,
then
every convergent subsequence tends to f. In this case (1.13) implies that the
sequence (f.) itself converges pointwise and f(z) = tim
locally uniformly
mA.
We remark that there is a certain analogy in the behavior of Schwarzian
derivatives and complex dilatations. In 1.4.6 we pointed out that locally
uniform convergence of quasiconformal mappings does not imply the convergence in
of their complex dilatations. On the other hand, Theorem 1.4.6 shows that the
of complex dilatations does imply
pointwise convergence of suitably normalized mappings.
1/
1.5. Area Theorem

Let us leave Schwarzian derivatives for a moment and establish a classical
result on univalent functions, which we shall need in estimating the norm of
the Schwarzian derivative of a conformal mapping
Theorem 1.2 (Area Tl/eorem). Letf be a univalent meromorphic function in the
domain {zJ Izi> 1}, with a power series expansion
f(z) = z +
(1.14)
Then
1.
(1.15)
The inequality is sharp.
PROOF. Let C, be the image of th circle
= p> 1 under f. The finite
domain bounded by C1, has the area

I
rn,, =
wdW.
w = f(z) and considering (1.14) we obtau

= np2
As
an area, rn,> 0 and the result (1.15) follows asp
1.
The Area theorem was first proved by Gronwall in 1914 and efficiently
used by Bieberbach two years later. It is a bistDrically significant result as it
59
i. Schwarzian Derivative
marks the beginning of the systematic theory of univalent functions. We need

the following immediate consequence of it:
Under the assumptions of the Area theorem,
(1.16)
1b11 1.
Equality holds
and only Vf(z) = z + eW/z.
For the extremal function of (1.16), the equality sign holds as well
(1.15).
For (1.15), there are many other extremals.
The functions satisfying the conditions of the Area theorem with b0 =0 are
said to form the class E. Another important class in the theory of univalent
functions is S. which consists of functions f univalent and holomorphic in
4.he unit disc with f(O) = 0, f'(O)= 1. If fcS and f(z) = z +
then
z . q,(z) = 1/f(z2)"2 belongs to E end tp(z) = z
a2
2,
with equality only for the Koebe functiQns z
z/(1 + e"z)2.
1ff eS and gis an arbitrary conformal self-mapping of the unit disc, then
fog f(g(O))
f'(g(O))g'(O)
belongs to S. Setting g(C) =

(1.18) and considering (1.17),
1(1
z)/(1 + IC), we obtain by calculating a2 for
4.
Integration of this inequality twice leads to the estimate If(z)I I

As IzI '
1.1:1.
+ zfl2.
we get the Koebe one-quarter theorem which we used already in
1.6. Conformal Mappings of a Disc

By Theorem 1.1, any function holomorphic in a simply connected domain
A of the complex plane is the Schwarzian derivative of a function f which is
meromorphic and locally injective in A. It follows that
can grow
arbitrarily rapidly as z tends to the boundary of A. In particular, there are

many classes of functions f for which the norm of S, is infinite. Even if f is
bounded, it may happen that II IL4 =
An example is the function
z -f(z) = exp((z + 1)/(z
1))
in the unit disc, for which S1(z) = 2(1

The behavior of (1.19) differs greatly from that of a Mbius transformation
in that (1.19) takes every value belonging to its range infinitely many times.
In contrast, a univalent f is analogous to a Mbius transformation in its
11. Univalent Functions
60
value distribution, and it will turn out that the norm of its Schwarzian is
always finite. We shall first prove this in the case where A is a disc, a disc
meaning a domain bounded by a circle or a straight line.
Theorem 1.3. 1ff is a conformal mapping of a disc, then
(1.20)
II S1 II 6.
The bound is sharp.
PROOF. By formula (1.12) it does not matter in which disc f is defined. We
suppose that f is a conformal mapping of the unit disc D. Let us choose a

By (1.9), this
point z0eD and estimate ISj(z0)I',(z012 = (1
expression is invariant under Mbius transformations. Hence, we may assume that = 0. Also, since f can be replaced by h of, where h is an arbitrary Mbius transformation, there is no loss of generality in supposing that
f S. Let denote the n th power series coefficient off
The function
z .l/f(t/z) =
the conditions of the Area theorem. From b1 = a3 we thus

conclude that
I. On the other hand, S1(0) 6(a3

Consequently, S1(0)I 6, and (1.20) follows.
For the Koebe functions f the coefficient b1 of z -+ 1/f(1/z) is of absolute
value 1. Hence, for the Koebe functions
a31 = 1, and equality holds in
(1.20). More generally, in D equality holds in (1.20) for all functions hofog,
where g is a conformal self-mapping of D, f a Koebe function, and h an
arbitrary Mbius transformation. In the upper half-plane, z ' f(z) = z2 is a
simple example of a univalent function for which II
= 6.
D
satisfies
The estimate (1.20) was proved by Kraus [1] in 1932. His paper was
forgotten and rediscovered only in the late sixties. Meanwhile, (1.20) was
attributed to Nehari ([1]) who proved it in 1949.

2.1. Distance from a Disc
Let A be an arbitrary simply connected domain which is conformally equi-
valent to a disc. Even in this general case, the norm of the Schwarzian
derivative of a conformal mapping of A is always finite, but the bound may
be as high as 12.
61
In order to study the situation more closely, we introduce the domain

constant
5(A) =
II Sf
where f is a conformal mapping of A onto a disc. We call 5(A) the distance of

the domain A from a disc. In view of the invariance of the norm of the
Schwarzian derivative under Mbius transformations, the distance 5(A) is

well defined.
We have 6(A) = 0 if and only if A is a disc. By Theorem 1.3 and formula

(1.1 1),
for all domains A. The distance 5(A) measures how much A deviates from a
disc, or equivalently, how much a function f mapping A onto a disc deviates
from a Mbius transformation.
An illustrative example is provided by the case in which A is the exterior of the ellipse {z =
k = 1, the
k
+
ellipse degenerates into the line segment with endpoints 2.) The function z -+f(z) = z + k/z maps the disc E = {zjIzI> 1) conformally onto A.
Because S1(z) = 6k(z2
k)" 2,
follows that
5(A) = 6k.
We see that 5(A) changes continuously from 0 to 6 as k increases from 0 to 1.

Thus the range of 5(A) for varying domains A is the closed interval [0,6].
Another simple example is the angular domain A {zIO < argz < kn},
0 < k 2. Now z ' f(z) = z" maps the upper half-plane onto A. From
S1(z) = (1 k2)/(2z2) we obtain
5(A)
II.
(2.1)
Again, 6(A) covers the whole closed interval from 0 to 6 as k grows from I to
2.
It might be expected that domains close to a disc are quasidiscs. This is

indeed the case: In section 3 we shall prove that for a K-quasidisc A the
distance 5(A) *0 as K -+ 1, and in section 5 that all domains with 5(A) < 2
are quasidiscs.
2.2. Distance Function and Coefficient Problems

The problem of estimating the distance function 5(A) is connected with
classical problems regarding the power series coefficients of univalent functions. Theorem 1.3 gave an indication of this. To make this more precise, we
first note that in view of formula (1.11), we could define 6(A) with the aid of
conformal mappings of the unit disc D onto A. Let g be a conformal selfmapping of D, such that g(0) =
Since
= 1, we conclude from the
62
invariance relation (1.9) that
ISf(zo)Ifl(Zo)2 =
This yields the characterization
b(A) = sup{ IS,(O)I If: D ' A conformal)
(2.2)
for the distance function.
Since composition of f with a Mbius transformation does not change

(), we may further assume that f(O) = 0, f'(O) = 1 and that cc is not in A,
i.e., thatfeS. Then

S,(O) = 6(a3
(2.3)
Hence, determining () amounts to maximizing the expression 1a3
2.3.
Boundary Rotation
As an example of the use of the characterization (2.2) and the formula (2.3),
we shall determine the sharp upper bound for (A) in certain classes of
domains characterized geometrically in terms of their boundary. We shall
first introduce "boundary rotation".
Let A be a domain of the complex plane whose boundary is a regular

Jordan curve. It follows that .3A is the image of the interval [0, 2ir) under a
continuously differentiable injection y with a non-zero derivative. Let
be the angle between the tangent vector y'(t) to 3A at y(t) and the vector in
the direction of the positive real axis. We assume that y(t) describes A in
the positive direction with respect to A as t increases. Then
diji(t) =
2.
(2.4)
The boundary rotation of A is the total variation of iriji. If

I
< 00,
(2.5)
Jo
the domain A is said to have the boundary rotation kit.
For an arbitrary domain A in the complex plane conformally equivalent to

a disc, boundary rotation is defined as follows. Let A,, n = 1, 2, ..., be an
exhaustion of A, i.e., A, A,1 A for every n and u A, = A. Suppose that
4,, B,,, B, A, and that B, is bounded by a regular Jordan curve. For a
rixed n, let a,, be the infimum of the boundary rotations of all such domains
B,. Then the boundary rotation of A is defined to be a = lima,,. The limit a
loes not depend on the choice of the exhausting domains A,,, and it agrees
with the pre iously defined boundary rotation for domains whose boundary
is a regular Jordan curve.
Boundary rotation can also be defined directly for the above domain A in
2.
between Simply Connected Domains
purely analytic terms. Let

A, and let
f be a conformal mapping of the unit disc D oni

u(z)
=1 +
Then the boundary rotation of A is equal to

lim
r
Iu(re'iIdq,.
Fi
Suppose that A has finite boundary rotation kit. From the finiteness of (2.7)
it follows that the harmonic function u can be represented by means of the
PoissonStieltjes formula. Integration then yields
f'(z) = f (O)e
105(1
dtp(8)
(2.8)
i/i is a function of bounded variation satisfying (2.4) and (2.5). It agrees

with the defined earlier using the tangent if 8A is a regular Jordan curve. In
integrated form (2.8)is a generalization of the classical SchwarzChristoffel
Here
formula for the function mapping a disc conformally onto the interior of a
polygon.
Conversely, let i,1' be a function of bounded variation satisfying (2.4) and
(2.5). Under the additional condition k 4, a function f whose derivative is
defined by (2.8) is then univalent in D and maps D onto a domain with

boundary rotation kn.
have finite boundary rotation.More exactly, a
Convex domains of
domain is convex if and only if its boundary rotation is 2it. Equivalent to this
is the assertion that the function u defined by (2.6) is positive in D or that the
function ,fr is non-decreasing.
For convex domains formula (2.8) was derived by Study in 1913. The
notion of boundary rotation was introduced in 1931 by Paatero; his. thesis
[1] contains detailed proofs for all the results in this subsection.
2.4. Domains of Bounded Boundary Rotation

Using the representation formula (2.8) we can easily estimate b(A) for convex
domains.
Theorem 2.1. If A is Mbius equivalent to a convex domain, then
2.
(2.9)
Equality holds if A is the image of a parallel strip under a Mbius

formation.
PROOF. We may assume
that A itself is convex. Let f be an arbitrary con-
formal mapping of D onto A. In view of (2.2), inequality (2.9) follows if we
64
we may replace f by the function z cf(ze1') for

2.
c complex and q real, there is no loes of generality in assuming that S,(0) 0
and that f'(O) = 1.
From (2.8) we obtain by direct computation
prove that 1S1(0)l
/ '2*
S1(O)
(2.10)
= Jo
JO
0, it follows that
Since S,(0) is real and
Jo
\Jo
/ '2*
=f
Jo
\Jo
\2
cosOd*(O))
+ I
Jo
(2.11)
sin2
j'2z
2.
Jo
Because S1(0) 0, we have proved (2.9).

Equality holds if
= 2,
Jo
Jo
cosOdqi(O)
= 0.
These conditions are fuffilled if i/i has a jump +1 at the points 0 and ir and is
constant on the intervals (0, it) and (it, 2ir). Then Sf(O) = 2, and it follows from
(2.8) that f'(z) = (1
22)1. We conclude that the image of D is a parallel
strip.
Theorem 2.1 expresses in a quantitative manner the fact that a convex

domain is close to a disc: Its distance to a disc is at most 2, while the distance
can be as large as 6 in The general case.
Not all domains close to a disc need be Mbius equivalent to a convex

domain, as the example A = (zIO <argz < kit), k> 1, shows. Its boundary
forms two interior angles > it, one at 0 and the other at
Since these angles
are preserved under Mbius transformations, A is not Mbius equivalent
to a convex domain. On the other hand, by formula (2.1) we have ()
2(k2
1).
Theorem 2.1 can be generalized.

Theorem 2.2. Let A be Mbius equivalent to a domain with boundary rotation
kx. If k 4, then
5(A)
The bound is sharp.
(2.12)
65
2.. Distance between Simply Connected Domains
k=3.5
k=2.S
Figure 4. Extremal domains.
The main lines of the proof are the same as those in Theorem 2.1. After
similar initial remarks we start from (2.10), assuming this time that S1(0) <0.
In the first line of (2.11) we now ignore the third integral and conclude that
/ r2s
\2
cosOdi/i(9))
f'2*
cos2Odi,140).
Jo
/ Jo
With attention paid to (2.4) and (2.5), the estimate (2.12) follows from this
I
after some computation; for the details we refer to Lebto and Tammi [1],
p. 255.
From (2.8) we get for the extremal f with f(O) = 0 the representation
f(z)
(1
(i +
The corresponding domain A =
4 2k1
6k
s\k14+1/2 d
f(D) is symmetric with respect to the real
axis. Its boundary Consists of two half-rays in the right half-plane emanating
from a point of the real axis and forming the angle knf2 in A, and of a
vertical line in the left half-plane (Fig. 4). As k grows from 2 to 4, the vertical
line moves to the left until It disappears when k = 4, i.e., it then reduces to the
point
2.5. Upper Estimate for the Schwarzian Derivative

The use of the distance function makes it possible to generalize Theorem
1.3 in a precise form.
Theorem 2.3. Let A and A' be domains conformally equivalent to a disc and
f: A + A' a conformal mapping. Then

IFS1 IA
+ b(A').
(2.13)
The estimate is sharpfor any given pair of domains A and A'.
PROOF. Let h be a conformal mapping of the unit disc D onto A. Froni
f = (foh)oh1 we conclude that IISf hA
= IS10,,
5511D Since
66
H ljnivaJent Functiona
II
'.
the triangle inequality yields (2.13).
In order to verify that the estimate (2.13) cannot be improved, we consider

conformal mappings h1: D -. A, h2: D A'. Given an s >0, we choose h1
and h2 such that
This is possible, because
is
tions (cf the reasoning in 2.2),
Let g be the rotation z -+
Theut
onto A', and
(2.14)
1, 2.
invariant under Mbius transforma-
maps A conformally
IIS,IIA=
Now
For a suitable 9 we obtain from this and (2.14),

> 5(A') + 5(A) 2s.
HS,IIA
Consequently, (2.13) is sharp.
From Theorem 2.3 we obtain the new characterization

5(A) =
S1
conformal self-mapping of A}
for the distance function.
2.6. Outer Radius of Univalence

Let us introduce another domain constant
c0(A) = sup(flS,IIA jf univalent in A}.
We call
the outer radius of univalence of A. In 111.5.1 we shall also define
the inner radius of univalence of A.
Theorem 2.3 shows that there is a simple connection between the outer
radius of univalence and the distance function S (Lehto [6]).
Theorem 2.4. Let A be a domain conformally equivalent to a disc. Then
a0(A)
5(A) + 6.
PROOF. We write the definition of o0(A) in the form

c0(A)
sup{IISJIIAI.f:
-. A' conformal}.
(2.15)
67
DMtancc bctwccn Simply Connccted Domains
Then it follows from Theorem 2.3 that

o0(A) = () + sup(A').
4,
Hence, we obtain (2.15) from Theorem 1.3.
We can thus extend Theorem 1.3 to domains A for which we know 6(A).
Forinstance, iffis univalcatinthesector A = {zfO < argz <kn), 1 k 2,

then uS, fl4 2k2 +4, or if f is univalent in a convex domain A, then
QS,1L4 8.
domains A and for all
Since 6(4) 6, we see tb.t in all simply

functions f univalent in A, we always have'
12.
The maximum 12 can be attained. We must then have 6(A) 6, and extremals are obtained, for instance, if we consider iuitable self-mappings of such
domains A.
As one example, let us consider the domain 4 which is the complement of
to the extended plane. In the
1
x 1) with
the line segment
discE= {wItwI> 1)
f
e"12hoh1.
a conformal self-mapping
Then every h, is a conformal mapping of E, and

of A. We have
From
_6eW/(w2
and
ItS,,
it follows that
If S1 HA
12
There is even a point of A, namely

value 12.
the function
at which
takes the
2.7. Distance between Arbitrary Domains

Let us consider again two domains A and A' which are conformally equivalent to a disc. As a generalization of the distance to a disc, we introduce the
number
(A,A')=inf{JIS,I)41f:A i A' conformal)
and call it the distance between the domains A and A'.
68
11. Univalent
The triangle inequality yields the estimates

15(A) S(A')p
S(A,A') 5() + 5(A').
Equality holds in both places if A or A' is a disc. A less trivial example of

equality in the lower estimate is provided by the angular domains
A = {zIO < argz <k,r},
o k
k' 2.Iff1(z) =
A' = {zlO <argz <k'ir},
zk,f2(z) = z*', then
ofr1 is a conformal mapping
of'A onto A'. Hence

S(A, A') It
If k'
S12
= 2(k'2
k2).
or if
1, we have 2(k'2k2)=2(1k2)---2(tk'2)
so that in these cases b(A,A') = 15(A) S(A')I.
If k> 1, k < 1, then 2(k'2 k2) = 5(A) + 5(A'), but .S(A, A') is presumably
I
(() c5(A')(,
smaller.
Let Cl. be the quotient of the set of domains conforinally equivalent to a

disc by the group of Mbius transformations. The equivalence class cOntaining all discs can be called the origin of i) and our previous function 5(A)
the distance to the origin of the equivalence class which contains A.
It is an open problem whether
S(A', A') =0
implies that A and A' are Mbius equivalent. If the answer is affirmative, then
(Cl, ) is a metric space. Another open problem is to determine the diameter
of Cl.
3. Conformal Mappings with Quasiconformal

Extensions
3.1. Deviation from Mbius Transformations
Let f be a sense-preserving
of the extended plane onto

1ff is
then f is a Mbius
dilatation ,z is small in absolute
mapping; we
almost like a
remind the reader of the
in 1.4.1.
geometric interpretation, of
Thus the number
can be regarded as a measure for the deviation off
from a Mbius transformation.
Suppose now that f is quasiconformal in the extended plane with complex
dilatation jt and that, furthermore, f is conformal in a simply connected
domain A. From the results in the preceding sections we know that there is
another measure for the deviation of f from a Mbius transformation, its
itself. 1ff is conformal
a quasiconformal mapping
value, we know that f behaves
69
3. Conformal Mappings with Quasiconformal Extensions
is small, then at least in A the

Schwarzian derivative: If the norm II
mapping f is
Mbius transformation.
is
norms
It turns out that
and IIS,IA are closely related. If II
i.e., the behavior of f in the coraplement of A, which
small, then so iS
contains the support of JI, is reflected m the behavior of f in A (Theorem 3.2
in this section). Conversely, 1ff is conformal in a quasidisc A and 11S11L4 is
small, then f can be extended to a quasiconformal mapping of the plane with
13
a small fl it II
II
(Tfieorem 4.1 in the following section).
The study of the relationships between complex dilatations and Schwarzian derivatives will be one of the leading themes of this monograph. Apart
from its intrinsic interest, the possibility of using these two apparently different measures of the deviation from Mbius transformations is important
in the theory of Teichmller spaces.
3.2. Dependence of a Mapping on its Complex Dilatation

In studying the effect of the complex dilatation on the Schwarzian derivative
we need a result detailing how a quasiconformal mapping changes when its
complex dilatation is multiplied by a complex number. We consider the

normalized case of Theorem 14.3 and make use of the representation formula
given therein.
Theorem 3.1. Let be a measurable function in the plane with bounded support
and
< 1. Let z -+ f(z, w) be the quasiconformal mapping of the plane with
complex dilatation
and with the property lim(f(z, w) z) 0 as z p x.
Then,for every fixed z
the function w + f(z, w) is holomorphic in the disc
[wf < l/F(uL.

Also,for every fixed z outside the support
function z -+ f(z, w) depend holomorphically on wfor
PROOF.
of the anal vfic
< I/
jJ j.t 33
By Theorem 1.4.3,
f(z)
f(z, 1) = z +
where we now write

instead of to accentuate the dependence 0!
it follows that
p. From the defii,ition of the functions
on
Hence,
f(z, w) =
+ >:
From formula (4.! i) Ifl 1.4.4 we see thit
converges uniformly
U. Univa'ent Functioni
70
< 1.
whenever
It follows that the power series (3.1) converges if
< 1. Consequently, w f(z, w) is analytic in the disc WI < 1/Il

Outside the support of p. the funbtion z ' f(z, w) is a conformal mapping.
is holomorphic, and
is no longer a
Also, each function z +
singular integral. Therefore, we can differentiate in (3.1) with respect to z term
I wf
(I
by term, without affecting the convergence of the series. If prime denotes

differentiation with respect to z, wc obtain
f'(z, w) = 1 +
and similarly for higher derivatives. It follows that all derivatives of z -. f(z, w)
depend holomorphically on w in the disc Iwl <
0
Theorem 3.1 makes it possible to study the dependence of the power series
coefficients of z ' f(z, w) on w. Let
f(z,w) =
in a neighborhood of infinity. Then the coefficients w *

are holomorphic
in Iwl <
To prove this we first note that if f is analytic for Izt > R, then z , f(Rz)/R
is in class Z. Therefore, we may assume without loss of generality that f

satisfies the conditions of the Area theorem.
First of all, we have
bin z(f(z, w)
z).
Here the convergence is uniform n w, because Schwarz's inequality and the

Area theorem yield the estimate
Iz(f(z,w)z)b1(w)j2
IzI
By Theorem 3.1, the function w -. z(f(z, w) z) is holomorphic ft4W1

for every fixed & Hence w ' b1(w) is bolomorphic, as the udiform
limit of holomorphic funCtions. The analyticity of w -. b,,(w) is deduced similarly from
lim za(f(z, w)
*.-I
z
b/w)zJ)
by induction.
The special normeftllon of the mappings is not essential in Theorem 3.1.
Corollary 3.t Let

upper
ping of the
a r4easurable function in the piane which vanishes in the

and jbr which flpIL, < 1. Letf,1,, be the quasiconformal map-s
p1mm. t.Mmmmplex dilatation
which keeps the points 0, 1, 00
71
3. Conformal Mappings with Quasiconlormal Extensions
fixed. Then the function

finite z.
for every
is holomorphic in wf < I p
w ,
PRooF. Let g be the Mbius transformation which maps the points 0, 1,

a quasiconformal mapping of the
on the points 1, 1, 1, respectively,
o g'. Then
plane whose complex dilatation wv agrees with that of
p(z) = v(g(z))g'(z)/g'(z).
satisfy the normalization condition

Further, let
By Theorem 3.1,
is analytic in
<
Set a1 =
1), a2 =
and
a3 =
h
a2
0 as
= l/II.uL.
a3 C a1
a2 a1
has the complex dilatation wp, and it fixes 0, 1,
Then
quently,
Conse-
=
depends analytically
By applying Theorem 3.1 again, we conclude that
is holomorphic in the disc claimed.
on w. It follows that w .-.
0
we also deduce that for every z in the upper halfdepend holomorphically on w in the disc
plane, the derivatives of z
Iwl <
From
h,.,, o
og
3.1 (and Corollary 3.1) can be generalized: The mapping

depends holomorphically on w if its complex dilatation p(, w) is a holomorphic function of w. We shall not need the result in this generality. (A proof,
Remark. Theorem
which is still a straightforward application of the representation formula

for f(z, w), is in Lehto [3].) For most of our applications, the simple case
w) = wp(z) is suflicient, but in V.5 we shall also be dealing with complex
dilatations of the form wp + v. In this case, the generalization of Theorem 3.1
is immediate.
More precisely, let us assume that p and v vanish outside of a disc and that
+ v) =
1 = 1, 2, ..., where P, is a
Hi'L < 1, If vL < 1. Now
polynomial in w of degree i. We again use the fact that
converges
<1. It follows that in the representation
uniformly whenever fi
f(z, wp + v) =
the right-hand series is uniformly convergent in w, provided that II wp +

< 1. We conclude that for every finite z, the function w + f(z, wp + v) is
holomorphic in the disc I <(I
By using this result we see that in Corollary 3.1, the complex dilatation wp
72
IL Univalent Functions
can be replaced by wp + v. Also, the holomorphic dependence of the derivatives on the parameter w remains in effect if wp is replaced by wjt + v.
Ahifors and Bers realized that in the theory of Teichmller spaces it is of
basic importance to study quasiconformaj mappings with varying complex
dilatations. In their joint paper [1] they proved the holomorphic dependence
of the mapping on its complex dilatation.
3.3. Schwarzian Derivatives and Complex Dilatations

Using Theorem 3.1, we can prove a result which shows that a small complex
dilatation forces the Schwarzian derivative to be small.
theorem 3.2. Let f be a quasiconfosmal manping of the

whch has the
complex dilatation p and which is conformal uz a simph' connected domain A
with at least two boundary points. Then
(3.2)
tSf}A IA
If q is a Mbius transformation, we Lan replace j by

without
changing the norms of either the Schwarzian derivative or the complex
PROOF.
dilatation. Also,
c A.
Then p has bounded support.
Let w be a complex number with Iwl < 1. We
for a moment the
unique quasiconforrna! mapping z f(z, will
of the plane which has the
complex dilatation
and the property 1(z.
z +0 as
z . cia. (We may assume that
> 0.) By Theorem 3.1, the
of
the analytic function z ' (f A)(z, w/ll /A
with respect to z depend analytically on w in the unit disc, at every finite point z of A.
Keeping z fixed, we define the function
II
w +
Sf(.,
Since S1 is a rational function of the first three derivatives of fI A, we conclude

that is analytic in the unit disc Iwt < I.
the function ii' is
bounded:
From the fact that : f(z.O! the identity mapping it follows that
= 0. We can therefore apply
lemma to if.'
and get
ao(A)iwl.
Setting w IIpL. we get back, modulo a
tion z --+ f(z) we started with, and (3.2) follows.
transformation, the fur.c-
If A is a disc, (3.2) assumes the form

IISIIAII
(Khnau il], Lehto [I]). The bound is sharp: For each
k,
0 k < 1,
73
3. Confonflei Mappings with Quasiconformal Extensions
there are mappings f for which

A = (zUzI>
1)
= 6k.
II
For
instance, in. tbe as
the function f defined by
f(z)=z+ki
if
if
tzI1,
(3.3)
is such an extremal. Other extrernals are obtained if this (is composed with
Mbhis tran*formatidns.
Using Theorem 3.2 we can estimate () for quasidiscs.
Theorem 33. If A is a K-quasidisc, then
K2t
(A)6K21.
(3.4)
H
under a K2-quaslconformal mapping! of the plane which is conforma) inK.
PROOF. By Lemma 1.6.2, the domain A is the image of the upper
By Theorem 3.2,
f
On the othcr hand,
11,1
= i.4).
It is not known whether the bound in (3.4) is sharp. From the example (3.3)
we deduce that the sharp bound is 6(K t)/(K + 1). In any event,
equality (3.4) shows that for a K-quasidisc 4, the distance () ' 0 as K ' 1.
3.4. Asymptotic. Estimates
Application of the representation formula (3.1) and reasoning similar to that
used in proving TheOrem 3.2 make it possible to obtain readily a number of
The rest of
results for conformi.t mappings with
this section will be levoted to questions of this type. This entails a

detour from
main then)e, the connedion between Scbwarzian denvatives
and complex thlatations.'
E; i.e., fis
mappings f which betong to
Let us consider
univalent in E = {zflzl > 1) and has power series expansion of
in E. 1ff has a quasiconformal extension to the plane with complex dilatation

p satisfying the inequality II a. k < I, we say that f belongs to the subclass
of . By abuse of notation, we sometimes use the symbol f both for the

conformal mapping of E and for its extension to the plane.
We shall first derive asymptotic estimates for
in
as
ri and
k
0.
ii. univaicut Functions
and k <k0 < I. Ask
Theorem 3.4. Letf E
f(z) = z
0,
if
+ 0(k2)
(3.6)
in the whole plane. Here I0(k2)I ck2, the constant c depending only
< 1, we seefrom formula (4.15) in 1.4.4 that
PROOF. If p> 2 and
ck2.
EITq,,(z)I
Hence, Theorem 3.4 follows from Theorem 1.4.3.
Corollary 3.2. Thefunctionsf E Ek satisfy the asymptotic inequality

(f(z) zi
RJJDR_ZI
=
K
zI
ck2.
a.e.,
(3.7)
(3.8)
then
lf(z) zi
2VJJDIC_ZI + 0(k2).
2k + ck2. The mapping with
For z = 0, the estimate (3.7) gives

(3.8) can be determined:
extremal
Jz
+ k2e2/z
+
if frI.> 1,
1)
+ k2e2i
if frI 1.
Hence (f(O)( = 2k. These functions I arc not only asymptotically extremal
but they actually maximize (1(0)1 in (Khnau (23, Lchto [3)).
For the coefficients b,, in (3.5), a counterpart of (3.7) can also be easily
established.
2k
IbAI41 +ck2,
with c
k)1.
(3.9)
If
f5(z)
n=1,2,...,
+
+
2k/(n +
1).
if zI> I,
if (zI 1,
3 10
75
with Quasiconformal Edsesless
PROOF.
We have
=
for IzI>
1.
Hence,
R,..1
(3.11)
,fJD
Mcbwarz's inequality and the estimate (4.13) in 1.4.4 for p = 2 (in which case
NH)1,=l)yield
ltfl
ffJI,
Consequently, we have the asymptot c representation
n = 1,2,...,
+ 0(k2),
=
the remainder term being. n"2k2/(l
I:) in absolute value. From this (3.9)
follows.
We conclude by easy computation that bN = 2k/(n + 1) + 0(k2) if

a.e.
Direct verification shows that the function f,, defined by (3.10) has this complex dilatation. We also see that for j, the Coefficient b1 = 2k/(n + 1).
0
Inequality (3.9) was established by Khnau [2] using variational methods.

Note that the function. (3.10) are related to each other by
f(z) =
1,
are also extremal. Some

the functions z ,
Along with
extremal domains arc pictured in
5.
In 3.6 we shall make a few more remarks on the coefficients b,, in
n1
na2
n 23
Figure 5. Extrcmal domains fork
76
ii. Univalent Functions
3.5. Majorant Principle

By the Area theorem, the coefficients of a function fe
satisfy the inequality
1. It follows that
=
f(z) zI
(3.12)
1.2.1 that is a normal family. The estimate

(3.12) also shows that E is closed, i.e., it contains the limits of its locally
We conclude from Theorem
uniformly convergent sequences.

Every subfamily
is then also normal. Applying Theorem 1.2.1 once more
extended quasiconformally to the plane with
we infer that the functions of
k, constitute a normal family. This implies that every
is closed
Hp1
also.
Let D be a complex-valued functional defined on

We say that .D is
.-*
continuous if
f(z) locally uniformly in E. For
whenever fM(z)
a continuous D, there are extremal functions which maximize
in Ek.
This follows immediately from the fact that Ek is a closed normal family. We
set
= max I'I'(f)l.
fEE,,
Clearly M(k) is non-decreasing in k.

Let us assume, in particular, that
is
a holomorphic function of finitely
many of the power series coefficients of f.and of the

of f and its
of E. We call such
at finitely many given
derivatives f', f", ...,
analytic. An analytic functional is continuous, because uniform
convergence of analytic functions implies uniform convergence of their derivatives. An example of an analytic functional is G(f) = (1z12 . I)2Sf(z),
considered in Theorem 3.2, which is. a rational function of f'(z), f"(z) and
f" (z).
Applying Schwarz's lemma as in the proof of Theorem 3.2, we get the

following general result (Lehto [3]).
Theorem 3.6. Let d be an analytic functional on E which vanishes for the

identity mapping. Then M(k)/k is non-decreasing on the interval (0, 1).
PROOF.
Fix k and k , 0
<k <k' < 1, and choose an arbitrary mapping f0 e
Let p he the complex dilatation of

extension off0, p
k. Consider
the mappings
which have ihe complex dilatation wj.z with IwI <k'/k.
By Theorem 3.1,
depends holomorphically on w in the disc
<k'/k.
if w 0, then f is the
Therefore, by Schwarz's 1cm
mapping, so that D(f) vanishes at w =
0.
77
3. Conformal Mappings with Quasiconforinal Extensions
For w = 1 we have f = Jo. Since is an arbitrary element of

desired inequality M(k) kM(k')/k'.
we get the
Is an analytic functional on
Corollary 33 (Majorant Principle).
0
which
vanishes for the identity mapping, then
kmaxib(f)I.
feZ
feE,
(3.13)
If equality holds for one value k e (0, 1), then it holds for all values of k.
PROOF. Inequality (3.13) follows ithmediately from Theorem 3.6 if we let
1.
Suppose that equality holds in (3.13) for some value k, 0 <k < 1. Let fk
dilatation of its extension. For
be extremal in this Ek and the
functions f with complex dilatation
Iwl < 1/k, Schwarz's lemma gives
over L But now equality holds
kM(1)IwI, where M(1) =
for w = 1. Ii follows that
= kM(1)IwI
in the whole disc
(3.14)
< 1/k. If k' e [0,1) is arbitrarily given, then for w = k'/k
the function f is in E'.. Combining (3.13) and (3.14) we deduce that I is

extremal in E'.. In other words, if equality holds in (3.13) for one value
is an extremal complex
k (0, 1), then it holds for all values of k, and if
are extremal.
dilatation, then all dilatations
<
U
3.6. Coefficient Estimates

Let us illustrate the majorant principle (3.13) with an example. We choose
(f) b5, for an arbitrary fixed n. This is admissible, because every is
zero for the identity mapping. It follows that
max)b1 kmaxlbj.
(3.15)
I
Assume that for a fixed n, equality holds for some k > 0. By what was said
about equality in (3.13), equality then holds for all values of k. It follows that
the extremal
for k is k/k0-times the extremal Ib,,I for k0. Furthermore,
is an extremal dilatation for k0, then wt0 is extremal for k with IwI = k/k0.
then is multiNow consider the formula (3.1 1). If is replaced by
plied by the power w'. It follows that all terms in the series (3.11) vanish,
except for the first one. But then
2k/(n + 1), and by (3.15),
2/(n + 1).
This is
to be true if n = 1, 2. (For classical results on univalent
functions we refer to Duren's monograph [1], which also contains an exten-
U. UakA
78
save
+1) for
Theorem 3A
Thus in
mm4b1l
in Ek, k
k(1/2 + e'). For a 4,
1,2.
In Z, maxlb,I
112 + 1',
that maxlb,I is strjctl)I a, than
whereas
unimown both in
apd' X,. Prom
Theorem 3.5 it
for all values of a.

Tbc Area t
requires a
for Z,. Gctting its sharp form

ottha reasoning which led to (3.13).
The estimate is shdrp.
PRooF. Given an arbitrary function leEk with the coefficients

we set
if b,,
other
Letu be thecomplexdilaletio of
the extended f, and b(w) the
of the function z-if(z,w/k) with
2:-normalization and complex dilatation w,s/k.
For an arbitrary positive integer N, we write
N
ii
is holoniorphic m the unit

The Area theorem gives
the estimate
1. Since b(O)=O, the functi9n
hssazero of order
2 at the origin. Schwarz's lemma therefore yields the improved timate.
By Theorem 3.1,
(3.17)
I*(w)I
If we set w =
we
getback thefunctionf with which we started. Hence

N
The desired inequality (3.16) follows as N -.
in Izi 1. A relatively simple argument shows that there are no other cxtremals (Lehto [1]).
The result (3.16) is due to Kuhnau [2] and Lehto [1].

In V.7.5 we shall again use an estimate of type (3.17) in connection with.
another problem where we know that the Jiolomorphic function under consideration has at least a double zero at the origin.
4. Univalence and Quasiconformal Extensibility of Meromorphic Functions
79
The general inequality (3.13) allows many other applications. For instance,
the classical Grunsky and Golusin inequalities in E can be immediately

It is also possible to change the original setting and use
sharpened for
other normalizations for the mappings. If we consider the class S instead of
E and define the subclass Sft of S in the same way we defined Ek, a difficulty
is encountered, because the complex dilatation of the extended mapping does
not determine the element of uniquely (S contains many Mbius transformations). Unique correspondence follows, for example, if we require that
the extended mappings fix
For this subclass Sk(
we can sharpen a
number of results known to be valid in S. (For various applications of the
majorant principle, see e.g. Lehto [3], [5].)
Univalent functions with quasiconformal extensions can also be studied by
use of variational techniques. Compared to (3.13), such methods often involve
much more laborious computations, but are essentially better in many cases
where (3.13) fails to give a sharp result. Khnau is a pioneer in this field. He
has had many successors, so that an extensive literature exists today. Delving
into these questions would take us too far afield from our main topic, and we
content ourselves here with mentioning, besides Khnau's works [1] and [2],
the papers of Schiffer [1] and SchifTerSchober [1], the lecture notes of
[1], and U e monographs of Pommerenke [1] and Krushkal
Khnau [1] among
othets.
4. Univalence and Quasiconformal Extensibility of

Meromorphic Functions
4.1. Quasiconformal Reflections under Mbius
Transformations
Theorem 3.2 establishes a relation between complex dilatations and Schwar-
zian derivatives in one direction: If a quasiconformal mapping of a plane

which is conformal in a simply connected domain A is close to a Mbius
transformation in the sense that fls complex dilatation is small, then it is close
to a Mbius transformation also in the sense that its Schwarzian drivative
is small in A.
We shall now establish a result in the opposite direction: 1ff is meromor-
phic in a quasidisc A and has a small Schwarzian derivative, then .f is

univalent and can be extended to a quasiconformal mapping of the plane
with a small complex dilatation. We shall then complement this result by
proving that the result does not hold for a simply connected domain A unless
A is a quasidisc.
The extension of f is carried out by means of smooth quasiconformal
80
Function5
U.
reflections. We. begin by studying the behavior of quasiconformal reflections
under conjugation by Mbius transformations.
Let C
aquasicircle boundMb the domains A1 and A2 and

a
K.quasiconformal reflection in C. If.h is a Mbius transformation, then
= ho ic o,lr' is a quasiconformal reflection n h(C). Using the identity
be
h(z1) h(z2)
z2),
valid for allMbius transformations, we shall establish two formulas, both of

which reveal invariance properties of quasiconformal reflections with respect
to Mbius transformations.
Writing C = h(z), Co = h(z0), we conclude from
= h(*(z)) that
Co
Zn). Hence
Co
(h'(*(z))'\112 *(z)
k h'(z)
CC0
Now fix Z0E C, such that 20, h(z0)
I
Then also
oo, and let z '
and we obtain our first invariance
linisup
If C passes through
Co
CC0
= lim sup
#(z)
Z0
(4.1)
then by formula (6.2) in 1.6.2
c1(K)
(4.2)
z-zo
with a finite constant c1(K) depending only on K. The invariance (4.1) shows
that inequality (4.2) holds for all K-quasiconformal reflections.
In order to derive the second invrriance property, we assume that

continuously differentiable off C. From
d.fr(z)/dz =
Combined with the identity q'(C)
is
(z) = dq(C) we then obtain
(0/dc.
(h'(,4r(z))h'(z))1P(*(z)
z),
this yields
Finally, let f be a locally injective meromorphic function in A2. If f =
Jo h1, then S1(li(z)) =

ance
(li(z) z)2
and we arrive at our second invari-
di/i(z)
d(p(C)
C)2
(4.3)
This invariance gives an estimate which will be needed in what follows:

Let C be a K-quasicircle bounding the domains
and A2. There exists a
quasiconformal reflection ,,li in C such that, for every function f meromorphic
and locally injective in A2,
$.Univalence ndQuuiconformal Exteniibility of Meromorphic Functioni
c(K)IIS,HA.
I*(z) zI2Id%fr(z)/dzl
81
(4.4)
oo). Here the finite constant c(K) depends
at each point ze A1, z

only on K.
If C passes through oo, We tke = q as in Lemma 1.6.4. Then (4.4):.

follows directly from the formulas (6.1) and (6.4) in 1.6. If C is bounded, we.
choose the Mbius transformation h so that e h(C). We then obtain (4.4)
from the invariance formula (4.3), because II S1
= uS1
0
4.2. Quasiconformal Extension of Conformal Mappings

We can now prove the main result on the role of the Scbwarzian derivative
for the univalence and quasiconformal extension of meromorphic functions
(Ahlfors [4]).
Theorem 4.1. Let A be a K-quasidisc. Then there is a constant c(K) > 0,

depending only on K, such that every function f meromorphic in A with the
property
US, hA <e(K)
(4.5)
is univalent in A and can be extended to a quasiconformal mapping of the plane

whose complex dilatation satisfies the inequality
(4.6)
Hi' II
PROOF. Let f be mereitorphic in A. We may assume that f is locally injective

does not lie in A. Let w1 and w2 be solutions of the differential
and that
equation w" + S,w/2 = 0 in A so normalized that w1 w2w'1 = 1. In
= 5,. It follows that I is the
proving Theorem 1.1 we showed that
composition of w1/w2 with a Mbius transformation. We may therefore take

I = wi/w2.
The desired extension off is obtained by an explicit construction. In order
to circumvent the difficulty arising from the fact that we have no a priori
knowledge about the behavior off on the boundary of A, we resort to an
approximation procedure. We assume first that w1, w2 and I are holomorphic on the boundary of A. Having proved the theorem under this additional
condition, we obtain the general result by exhausting A with subdomains
which are K-quasidiscs and on whose boundaries f has no poles.
Let A1 denote the complement of the closure of A and a quasiconformal
reflection satisfying (4.4); the domain A now, plays the role of A2 in the earlier
discussion. We write z =
and define a function g in A1 by
+ (C
W2(Z) +
82
II. Univalent
We prove
At infinity, g is defined as its limit: g(cx)) =
desired
continuation
off
if
(4.5)
holds
for
an
appropriate
4K)
that g is
It is immediate that o(C) -. f(z0) as approaches a boundary point eM.
Moreover, if is a finite point of A1 and u(C) 00, then g a continuIn view of the relations w1w
= 1 and
ously differentiable at
1,2, we obtain by direct computation
w1" =
1
8g(C
g(C)
+ (C z)w(z))2
S,(z)(C
With
From this we get an estimate for the complex dilatation p =
c(K) the constant in (4.4), we choose s(K) = 1/c(K). It then follows from (4.4)
that
II
S1H4
<1.
(4.8)
#0. Thus the Jacobian

We see from (4.7) that with this e(K), we have
= 18912(1
is positive at Hence g is injective at every finite point C
oo. By symmetry, 1/g is injective at every finite point of A1
at which
which is not a zero of g. Consequently, g is injective at all finite points of A1.
By considering the function C . g(1IC) we conclude that g is locally injective
at
also, and hence throughout A1.
Define a function F by F(z) = f(z) in A u t3A, and F(z) g(z) in A1.

Then F is iocally injective in A and in A1. Since F(z) 00 On 8A, the set
E = {zlF(z) = x,} consists of only finitely many points.
Let z0 be an arbitrary point of OA. Considering (4.2) we deduce by direct
computation that
f(z0)
= f'(z0).
Czo
sO
Applying (4.4) for z f(z) = log(z
we conclude that (C
f.'(z0)
Consequently,
by
(4.7),
g(O
,
hd
,
z0.
sO
as
C
and (C
We
conclude
that
F
is
continuously
differentiable
on
8A,
g(C) .0 as C
z0.
and hence everywhere outside E. On 8A we have J,(z) = If'(z)12, and so F is
locally injective throughout the plane.
By the existence theorem for Beltrami equations (Theorem 1.4.4), there is a
quasiconformal homeomorphism w of the plane which has the same complex
satisfies the CaucbyEiemann
F a.e. The function p = F o
has L2-derivatives.
equation.
a.e.
As
a
quasiconformal
mapping
0
of F are continuous outside E, we conclude that outside
Since th
the function p has L2-derivatives. By the remark made
the
in
with formula (4.5) in 1.4.3, sp is analytic in the complement
of
p is continuous everywhere, the points of w(E) cannot
be essential singularities, and it follows that q is meromorphic throughout
4. Univalenoe and Quasiconforinal Extensibility of Meromorphic Functions
83
the extended plane. Hence, .p is rational, and being locally injective, it is a

Mbius transformation. We conclude that F = ow is a quasiconformal
homeomorphism of the plane. This proves the theorem, provided that I is
holomorphic on the boundary of A.
4.3. Exhaustion by Quasidiscs

in order to complete the proof, we shall nowshow how to relax the restrictive
assumption that f is holomorphic on
Since A is a K-quasidisc, there
exists a K-quasiconformal mapping of the plane under which the unit disc
maps onto A. It follows that there is an increasing sequence of positive

numbers r,, < 1, n = 1, 2, ..., tending to 1, such thatthe image of each circle
= isa K-quasicircle on whichf(z) #
Let A1,, and A,, A denote the
complementary K-quasidiscs bounded by this quasicircie. As n 'cx, the
domains A,, exhaust A.
We proved in 1.1.1 that the Poincar density is smaller in A than in its
subdomain A,,. Therefore,
fw(
(4.9)
IS, hA.
It follows that condition (4.5) can be applied to f IA,. If we choose c(K) as in

the above proof (note that e(K) depends only on K, not on A), we conclude
that f IA,, agrees with the restriction to A,, of a quasiconformal mapping F,,
of the plane.
Let p,, be the complex dilatation of F,,. By (4.8) and (4.9)
lISf hA
2 c(K)IISfIIA
e(K)
<
We see that the maximal dilatations of the quasiconformal mappings F,, are
uniformly bounded. Thus these mappings constitute a normal family. The
limit of a locally uniformly convergent subsequence is a quasiconforinal
mapping of the plane whose restriction to A is f. Its complex dilatation also
satisfies (4.8), i.e., (4.6) is true and the theorem is proved.
0
4.4. Definition of Schwarzian Domains

Theorem 4.1 leads to the question whether the condition that A be a quasidisc
is necessary for the conclusion that a meromorphic function with a small

Schwarzian derivative is univalent in A and has a quasiconformal extension.
Or we may restrict ourselves to classical complex analysis and pose the

simpler question: In which domains does a small Schwarzian derivative
imply univalence?
Let us introduce the following definition: A simply connected domain A
with more than one boundary point is called a Schwarzian domain if there is
84
a positive constant such that every meromorphic function f with
I S1fl A
is univalent in A. More precisely, we say that A is then an e-Schwarzian

domain. This notion is due to Gehring [2].
Theorem 4.1 tells us that every quasidisc is a Schwarzian domain. It was
for a long time an open problem whether or not this sufficient condition was
also necessary, and few conjectures were expressed favoring either direction.
Finally, in 1977, Gehring [2] solved the problem by proving that every
Schwarzian domain is a quasidisc. Thus a small Scbwarzian derivative forces
a function to be univalent in A if and only if A is a quasidisc. This is one of
the unexpected characterizations of quasidiscs in terms of analysis we were
talking about in 1.6.7. Unexpected, because it shows that quasiconformal

mappings are intrinsic to the problem of relating the injectivity of meroinorphic functions to their Schwarzian derivative, a problem which on the sur
face has nothing to do with quasiconformality.
We shall now present Gehring's proof, which is based on the fact that a
linearly locally connected domain conforinally equivalent to a disc is a quasidisc (cf. 1.6.4 and 1.6.7).
4.5. Domains Not Linearly Locally Connected

Topological properties of plane domains can often be expressed in analytic
terms with the aid of the complex logarithm. We shall derive a result of this
type for domains which fail to be linearly localiy connected with a given
constant. (For the definition of linear local connectivity, see 1.6.4.)
Lemma 4.1. Let A be a simply connected domain which is not linearly locally
connected for a constant c> 1. Then there are two points z1, z2 of A and two
finite pouzts w1, w2 outside A, such that the function z + h(z) = log((z w1)/
(z w2)) satisfies the inequality
Ih(z1) h(z2)
21a1
(4.10)
PROOF. It follows from the assumption and from the definition of linear local
Either
connectivity that there is a disc D(z0, r) with the
there_are two points Pi and P2 lfl A D(z0, r) which cannot be joined in
A
D(z0, cr), or else there are two points in A\D(z0, r) which cannot be
joined in A\D(z0, rfc). Suppose initially that the former alternative is true.
Consider the line segment with endpoints Pt' P2 and a simple arc in A from
to P2 which meets the line segment at frnitely many points only. Among
these points of intersection there are two adjacent, 21 and z2, which cannot
be joined in A D(z0, cr). Let be the line segment with endpoints z1,
and
/1 the subarc from 22 to of the arc joining Pj and p2. Then ufl bwids
two Jordan domains A1 and A2; we assume that infinity lies in A2.
____.
85
4. Univalence and Quasiconforznai Extensibility of Meromorpbic Functions
For brevity we write D(z0, Cr) = U. Suppose that there are points w1 E
A,, i = 1, 2, in the complement CA of A. Then
h(z,) h(z2)
=J
((z w1)1
j((z wi)'
= 2iti(n(w1)
(z
Since w1 e A1,
'
(4.11)
(c 1)r
a c D(z0, r) and w1, w2 e D(z0, cr), we see that Iz

for every z a. Therefore
Because
f(
J,,\jzW1)
n(w2))
(z
where n(w1) is the winding number of a u fi with respect to

w2
have n(w1) = n 1, n(w2) 0. It follows that
Ih(z1)h(z2)2nnil
t3U
ci
Jzw21J
If n = 1, we thus get (4.1O) from (4.11) and (4.12). If is
1,
we obtain (4.10)
by interchanging the roles of w1 and w2.

We still have to prove that it is possible to find points w1 E .3U
are not in A, i.e., that
i=1,2.
(4.12)
A1 which
(4.13)
This requires topological arguments.

We first observe that, because a U and fi has points outside the closure
of U, the curve a u $ meets U in at least two points. By Kerkjrt's

theorem (Newman [1], p. 168), each component of the complement of
a u u au is a Jordan domain. In particular, each component of A1 n 11 is
a Jordan domain. Since a Jordan domain is locally connected at every bound-
ary point, there is a neighborhood V of z1 such that each pair of points in

A1
V can be joined in A1 U. Let ZEA1 n V and let A'1 be the z-component
of A1 U. Every point p of A1 Vcan be joined with z in A1 U, and the
joining arc lies in A'1. It follows that p E A'1, and so A1 V A'1 V, i.e., A'1
is the only component of A1 n U whose boundary contains z1.
If z'ci, we can join z' and z1 by an arc whose inner points lie in A1 U.
Then this arc is in the closure of A'1. Hence z'
A'1,
and we conclude that
a c 0A',. If y is the complement of a with respect to

then y c Au
r. A1). On the other hand, y joins z1 and in the closure of U. Thus y is
not contained in A, and (4.13) follows for i = 1. Similar reasoning yields (4.13)
for i = 2.
Finally, we have to consider the case in which there are two points in
A\D(z0, r) which cannot be joined in A\D(z0, nc). Let 1(z) = z0 + 1/(z z0).
By what we just proved, there are points
C2 in f(A) and points t1, t2
outside f(A), such that the function z - g(z) log((z t1 )/(z t,)) satisfies
86
the inequality
2iriI
(4.14)
= f'(t1), I = 1, 2, then direct calculation yields
If =
h(z) = g(f(z)) +
10gZO W1
zo w2
Hence, (4.10) follows from (4.14).
El
4.6. Schwarzian Domains and Quasidiscs

The result we set out to prove can now be established without difficulty.
Theorem 4.2. A Schwarzian domain is a quasidisc.
PROOF.
for a'
Let A be an a-Schwarzian domain. Then A is trivially d-Schwarzian

a. We may suppose, therefore, that a 2. (In 111.5 we shall show
that, in fact, no domain A is a-Schwarzian for a> 2, but here this result is not
does not lie
needed.) Nor is there any loss of generality in assuming that
in A.
We shall show that A is linearly locally connected with constant

c
= I + 16/a.
(4.15)
The theorem then follows from Theorems 1.6.5 and 1.6.6. More precisely, we
conclude that an a-Schwarzian domain is a K(a)..quasidisc, wheTe the constant K(a) depends only on a.
The proof is indirect. Suppose that A is not linearly locally connected with
the constant c = I + 16/a. By Lemma 4.1, there are points z1, z2 in A and w1,
w2 outside A, such that (4.10) holds. Clearly h(z1) h(z2).
Define
f(z)
Then f(zj)/f(z2) = I, so tbatf is not univalent.

From S1 = b2h'2/2 + Sh we get by an easy computation
Sf(z)
If
1 b2(
w3
w2
W2))
denotes the Poincar density of A, t;ien formula (1.5) in 1.1.1
the
estimate
(4.16)
8jb2
Since a 2, we see from (4.15) that c

2ir 1/2,
and so
8. Then Ih(z1)
h(z2)I
5. Functions
87
lent in a Disc
lbli
4*
4
1/2) < 5(c
It follows that
21
1b21i<
lO5(cl)
a
8
<a. Since A is an
domain,
This
is
a
contradiction,
and
so
A
is
linearly
locally
connected
f
with the constant c = I + 16/a.
0
We conclude from (4.16) that iJS1
is
The effct of the Schwarzian derivative on the univalence and 4uasiconformal ext"nsion of meromorphic functions will be studied in More detail in
sections 111.4 and 111.5.
5. Fun'ctions Univalent in a Disc

5.1. Quasiconformal Extension to the Complement of a Disc
Certain special curves admit simple quasiconformal reflections. In such cases
Theorem 4.1 can be expressed in a more explict form. Particularly imporlant
is the case where the domain is a disc. From the expression for the complex
dilatation of the extended mapping in this special setting we shall draw in

of Teichmuller spaces.
Chapter V important conclusions regarding the
Theorem 5.1. Let f be meromorphic in a disc D. If
fl
then
Sf
2,
f is univalent and can be extended to a quasiconformal mapping of the
plane. The constant 2 is best possible.

If D is the unit disc, there is. an extension with complex dilatation
1z12)2S1(z)
(5.1)
for z in D, while if D is the upper half-plane, an extensionexists with
--2y2Sy(z)
(5.2)
for z in D.
PRooF. Suppose first that D is the unit disc. In this case we have th simple
quasiconformal reflection C I/c. More precisely, in proving Theorem 4.1
we use in the approximation stage of the argument the reflection
=
< 1. By letting + 1, we then obtain Trom (4.7) the expression (5.1) for the
88
complex dilatation p
Since II
= 11sf 11/2, it follows that we can
take
= 2.
After this it is clear that
= 2 (= e(1)) will do for any disc, because all
discs are Mbius equivalent.

*. In this
If the domain is the upper half-plane, we have the reflection
case (4.7) yields the expression (5.2) for p.
In order to show that the bound 2 cannot be replaced by any larger, we

consider the function z + f(z) = log z in the upper half-plane H. This function f is univalent, and from S,(z) = (2z2)1 it follows that JIS,II = sup2y2/
which is
1z12 = 2. But the image of H is the parallel strip {wIO < 1mw <
not a Jordan domain. Consequently, f does not even possess a homeomorphic extension to the plane.
0
Here we obtained Theorem 5.1 as a corollary of the general Theorem 4.1,
but actually Theorem 5.1, proved by Ahlfors and Weill [1] in 1962, was
discovered before Theorem 4.1. Thanks to the simple reflections z , 1/2 for
the unit disc and z ' I for the half-plane, the proof of Theorem 5.1 is considerably shorter than that of Theorem 4.1. We can bypass the considerations
which guarantee the existence of the special reflection needed in the case of
an arbitrary quasidisc.
Theorem 5.1 gives the result we mentioned in 2.1:
Let A be a simply connected domain with more than one boundary point. if
the distance (A) from a disc is <2, then A is a quasidisc.
Let us return to (5.2) and set

half-plane
We then have in the lower
is a
function and the Poincar density.
Supplementing Theorem 4.1, Bers proved that the same result holds even
in the general case: A functionf which is meromorphic in a quasidisc A and has
where
a sufficiently small uS1 is univalent and has a quasiconformal extension with a

complex dilatation
where qi is a holomorphic function and
Poincar
density in the complement of the closure of
The complex dilatations
turn out to be important in Chapter V when
we consider quasiconformal mappings which are lifts to the universal covering surface of quasiconformal mappings between Riemann surfaces. In our
applications we shall get along with the case of the half-plane and content
ourselves therefore to indicating only briefly how the reasoning goes in the
general case. For the complete proof we refer to Bers [7] or [9].
Let M be the set of complex dilatations which are zero in A and of the form
outside the closure of A, and Q the space of holomorphic functions of A
with a finite hyperbolic sup-norm. If f is a quasiconformal mapping with
complex dilatation p in M, then S1 is uniquely determined by p. By methods
89
S. Ftmctiona Univalcnt in a Di,c
of functional analysis, Ben proves that the mapping z -, SM of M into Q is

invertible in a neighborhood of the origin of Q. The proof makes use of the
representation formula for of the reproducing property of the Bergman
kernel function and, just as in Theorem 4.1, of a Lipschitz-continuous quasiconformal reflection.
5.2. Real Analytic Solutions of the Boundary Value Problem

Interrupting briefly the study of functions meromorphic in a disc, we show
how Theorem 5.1 can be utilized for constructing real analytic solutions of
the boundary value probkm discussed in 1.5.
Let a k-quasisymmetric function h be given (cf. 1.5.2). We assume first that
k <,/i Let be a quasiconformal self-mapping of the upper half-plane H
with boundary values I.. We construct by using the BeurlingAhifors
1.5.3).
method so that the maximal dilatation of does not exceed k2
Bee use k2 <.2, the complex dilatation p of Ii satisfies the condition
$isL <
(5.3)
1/3.
By ThcoiemI.4.4, there exists a quasiconformal mapping f of the plane

which has the complex dilatation p in H and which is conformal in the lower
half.plane H'. By Theorem 3.2 and formula (5.3),
<2.
Thus we can apply Theorem 5.1 and obtain a new quasiconformal extension
of is a quasiconformal seif-niapping of H
f for flU'. Then
with boundary values h. From the expression
zEH', we see that f is real analytic. By the Uniqueness theorem 1.4.2, the
is conformal. Hence f* is real analytic.
mapping
In the case of an arbitrary quasisymmetric function h we write h =

where each h1 is k-quasisymmetric for k
Using formula
1,2
(417) in L4.7 we remark that this is possible, because the boundary function
is A(K)-quasisymmetric
of a K-quuiconformal self-mapping of H fixing
and .%(K) -. 1 as K -+ I. If f1 is a real analytic solution corresponding to
then f7 o o is a real analytic solution for h. We have thus given a proof
hR o'
for Theorem 1.5.3.
5.3. Criterion for Univalence

Theorem 1.3 says that if f is univalent in a disc, then IIS,fl
Theorem 5.1, we obtain a converse to this theorem.
6. By use of
90
II. Univalcnt Functions
Theorem 5.2. Letf be meromorphic in a disc. If
IIS,II2,
then f is
The bound 2 Is best possible.
n = 1, 2 ..., which are meromorphic in the

PROOF. Consider functions
given disc, fix three points of the disc, and have Schwarzians (1
by
Theorem 1.1 such functions exist Since II
<2, every is univalent owing
to-Theorem 5.1. They form a normal family, and the limit of a locally
uniformly convergent subsequence is a univalent function. Since the limit
function shares the same Schwarzian derivative with f, we conclude that f is
univalent.
In order to prove that the bound 2 cannot be replaced by a larger number,
consider the analytic function z ' f(z) = z1', s >0, in The upper hall-plane..
Then Sf(z) = (1 + e2)(2z2)', and so II S1 If = 2(1 + 82). On the other hand, I
is not univalent for any e> 0 For instance, f takes the same value at the
points i and I exp(2ir/s) of the upper half-plane.
Here we derived Theorem 5.2 as an easy corollary of Theorem 5.1. However, as might be expected, quasiconformal mappings are not needed for the
proof of Theorem 5.2, which is in fact a'much older result than Theorem 5.1.
It was proved in 1949 by Nehari [1], and in the same year Hue showed that
the bound 2 is sharp.
5.4. Parallel Strips

We shall now show that the condition If S1f1 2 n disc allows for conclusions about the function f beyond its univalenoc.
results in tho remaining
part of this section are due to Gebring and
[1].
Let f be a meromorphic function in a disc D, h a confonnal mapping of a
domain A onto D, and g fo h. Since fJ
II S9 Skfl .4, we can transfer
Theorem 5.2 to A. If we can construct h
modified theorem can
be of interest. In particular, if A is a parallel strip, certain tethnical advantages are gamed which will be utilized in the following.
Let us normalize the domains and assume that D is the unit disc and
A = (zI$ImzI <it/2}. Then
z ' h(z)
= tanh(z/2)
is a conformal map of A onto D. It maps the real axis R onto the real
diameter of D, and it has the Schwarzian Sh(z) = 1/2. The Poincar density
of A at z = x + iy has the value (2 cos
We list two immediate consequences. First, Theorem 5.2 assumes the following form: Let g be meromorphic in A = {zII!mzl <n/2}. If
91
2cos2y'
then g is univalent in A. The bound is best possible.

Second, if x is a point of the real axis and w = h(x), then by the basic
invariance formula (1.9) in 1.3,
41S9(x) +
= (1
1w12)21S1(w)I.
(5.4)
This equation remains valid if h is replaced by an arbitrary conformal map-'

ping of A onto D. The image of the real axis is always a geodesic line in the
Poincar metric of the unit disc, i.e., it is a circular arc which intersects the
unit circle orthogonally.
If f has no poles in D, then g is holomorphic in A. Direct computation
shows that on the real axis R, the function x . v(x) =
the differential equation
then satisfies
(5.5)
v"
with
4,
= Re S9
We remark that equation (5.5) with
Im
v-).
(5.6)
determined by (5.6) is an identity which
holds on R for every g holomorphic in A. Note that by (5.4), condition.

ItS1
2 implies Re S1(x) 0 and hence 4)(x) 0.
5.5. Continuous Extension

We shall prove that if flS, lID 2, then f always has a continuous extension
to the boundary of D. Using (5.5) and (5.6), we first establish as a preparatory
result an estimate on If'I when f satisfies certain additional conditions.
Lemma 5.1. Let f be
function meromorphic in the unit disc D satisfying
f"(O) = 0, IIS1 'ID 2, and (1 IwI2)21S1(w)I I in a neighborhood Iwl a < 1

of the origin. Then I is holomorphic in D and
tf'(w)I
/
i
I+
1
Iwl)
(5.7)
where M is a constant depending only on a.
By symmetry, it is sufficient to consider the case w > 0. Assume first

that f is holomorphic in D.
D defined in 5.4, the comWe make use of the conformal mapping h: A
Because
position g = foh, and the function x v(x) =
=0,
PRooF.
II.
92
we conclude from the assumption f"(O) =0 that g"(O) = 0. Hence

v'(O)
=0.
The hypothesis flS1 fl 1) 2 implies, as we mention&1 at the end of 5.4, that

q(x) 0 on R. By (5.5) we then have v"(x) 0, so that v' is increasing in x.
From v'(O) =0 it follows that v'(x) 0 for x 0. Consequently, v(x)
and by (5.5), v"(x)
From (1 1w12)21S,(w)t 1 in Iwl a we infer, in view of(5.4) and the
relation z = h1(w) = log((1 + w)/(1 w)), that ReS,(x) 1/4 if 0 x
b = Iog((l + a)/(1 a)). Hence, by (5.6), p(x) 1/8, and so v"(x)
v(O)/8
for 0 x b. This yields v'(x) xv(0)/8 for 0 x b. We conclude that
v(x)
for x
b.
v(O)(l + b(x
If c = min(l/b, b/8), it
b)/8)
follows that v(x)
cv(O)x, i.e.,
0. Because
= 21h'(x)f'(w)I = (1 1w12)1f1(w)I, 2g'(O) = f'(O),
x = Iog((l + w)/(1 w)), this is (5.7) with M =
Now drop the assumption that f is holomorphic in P and set r = inf
for
and
{IwoI w0 a pole of 1). Repetition of the above reasoning shows that (5.7)
then holds in Iwl <r. We see that if r < 1 and IwoI = r, then f'(w)l remains
bounded as w
' w0.
This is a contradiction, and so f has no poles in D. 0
Inequality (5.7) yfelds the following preliminary result: A function f

ing the conditions of Lemma 5.1 has a continuous extension to the boundary
of D. This follows from the fact that
+r)_2
is
integrable over the interval [a, 1].
Hence, we see from
f(r1e')I
that the limit of f(re') as r -+
exists
u/s(r)dr,
r2 > r1,
J 'I
uniformly in 0. It
defines,
therefore, a
continuous extension of f.
5.6. Image of Discs

We can now prove the main theorem about the image of a disc under a
conformal mapping f for which IS, II.
renke [1]) complements Theorem 5.2.
2. The result (Gehring and Pomme-
93
in a Dc
5. F*nclio
ThMrsm 5.3. Let fbe meronsorphic in a disc D and 115,11 2. Thenf(D)

is either a Jordan domain or the image of a parallel strip under a Mbius
transformation.
We may assaw.e that D is the unit disc. We first hote that
m = inf (1
(w12)21S,(w)I
= 0.
(5.8)
WED
the iaction i/Sf is holomorphic in D. From
For if in
(1
and
maximum pri*ciplc we then arrive at the impossible conclusion that
1/S1 varnshes i4entically in D.
F,om (51)
condude the existence of a point w0 D at which

IS,0V0)( <(1 IwoI2)_2.
that f(w0) =
We may
0.
If g1(w) = (w + w0)/(1 +
2and
IS,,(O)I = fS,(w0)I(1
and 11 =
jw012)2 < 1.
Thus there is an a >0 such that
<
<a. Next set c
(1 1w12)_2
w/(cw + 1), and 12 = g2 oft.

so
that
12
fulfills
all conditions of LemThen
and
= 0,
and hence
ma 5.1. It follows that 12 has a continuous extension to
of2ogr1 also has this property.
f
If the extended f is injective in the closure of D, then it is a homeomorfor
fID1(0)/2f(O)2, g2(w)
phism of the closure of D onto its image. In this case f(D) is a Jordan domain.
Suppose then that the extended! is not injective. Let f take the same value
at the points w1 and w2 in the closure of D. Since the condition S,ll 2

implies that f is injective in D, it follows that w1 and w2 lie on 3D. Let y be
the noneudidean line of D with endpoints w1 and w2, and w0 a fixed interior
point of y. We assume first that the value the extended I takes at w1 and w2
is
Then
the images of the components of y\{w0} under I both have
infinite length.
Let h now be a conformal mapping of the parallel strip A onto D which

maps the real axis onto y, and g = fo h. Then the g-images of the components
of R\{x0}, x0 = h'(w0), are the same as thef-images of y\(w0). Hence,
1
=
Jx0
On the other hand, we get upper estimates for these integrals by studying
the function v =
As before, we deduce from (5.5) that v' is increasing
IL Univalent Functions
94
in x. If v'(x0)>
0,
it follows that v(x) v(x0) + v'(x0)(x x0) for x
x0.
Hence,
I
v(x0)v (x0)
Jxo
Similarly, if v'(x0)
<0,
Ig'(x)I dx
In conjunction with (5.9) these estimates show that v'(x0) = 0, i.e., that v'
vanishes on the real axis. It follows that Re(g"/g') =
By
2v'/v 0 on
(5.6) and (5.5),
p=
0.
We conclude that g" vanishes on R and hence in A. Thus g is a similarity

transformation, and f(D) = g(A) is a parallel strip.
If the extended f takes a finite value c at w1 and w2, we can apply the above
reasoning to the function f1 = 1/(f c). We infer that f1(D) is a parallel strip,
and the proof of the theorem is completed.
0
5.7. Homeomorphic Extension

The following result (GehringPommercnke [I]), an easy consequence of
Theorem 5.3, fits in the narrow gap between Theorems 5.1 and 5.2.
Theorem 5.4. Let f be meromorphic and satisfy
IS,(z)I
In
<
(1 )z12)2
the unit disc. Thenf is univalent and has a homeomorphic extension to the
The image f(D) is a Jordan domain if

P100F. By Theorem 5.2, f is
and only 1ff has a homeomorphic extension to the plane. Hence, if a homeo.
morphic extension does not exist, then by Theorem 5.3, f(D) is the image of
the parallel strip A under a Mbius transformation. If h again denotes the

tanh(z/2) of A onto D, then g = fh.is a Mbius
conformal
It follows from (5.4) that

(1
at every point of Is(R). This is in
lz(2)21S,(z)I
with the hypothesis.
we obtain the followia pad.. classification for functions f

in the emit dlac CondIdo"
95
(1
z(2)21S1(z)l 2
implies that f is univalent, condition

(1
<2
that f is univalent and has a homeomorphic extension to the plane, and
condition
sup(1 )zV)2 jS1(z)J < 2
that f is univalent
has a quasiconformal extension to the plane.
CHAPTER HI
Universal Teichmller Space
Introduction to Chapter III

The notion of universal Teichmller space was crystallized in connection
with the problem of imbedding the Teichmlle space of a Riemana surface
into a space of Schwarzian derivatives. In the general case, the Schwarzians
in question are holomorphic quadratic differentials for a group of Mbius'
transformations (see V.4). The universal Teichmller space corresponds to
the situation in which the group is trivial. The Schwarzians are then just
holomorphic functions, and the machinery developed in Chapter II can be
applied directly. It follows that Chapter III, devoted to the study of the
universal Teichmller space, provides a bridge between univalent functions
and Teichmller spaces.
The other end of the bridge will not be visible until Teichmller spaces of
Riemann surfaces are introduced in Chapter V. The drawback is that in

section 1 of this chapter we are unable to motivate the definition of the
universal Teichmller space. In fact, plenty of explanation is required before
the role of the universal space in Teichmller theory becomes clear; a reader
who wishes to get an early idea of this role may want to consult V.3.
In section 1, various models of the universal Teichmi'ller space T are introduced, and the group structure of T is discussed.
Following Teichmller's classical example, we define in section 2 a distance
function which makes T a metric space. This space is shown to be pathwise
connected and
In section 3, we study the model of T provided by the family of normalized
quasisymmetric functions. This characterization offers certain technical advantages. Using it, we prove that T is contractible and that T is not a topological group.
1. ModeLs of the Universal Teichmuller Space
97
Section 4 deals with the problem of mapping T into the Banach space of
Schwarzian derivatives with finite norm. By appealing to the results of Chapter 11, we prove that the mapping is a homeomorphism of T onto its image
and that the image agrees with the interior of the set consisting of the
Schwarzians of univalent functions. The mapping could also be used to intro-
duce a natural complex analytic structure into the universal Teichmiiller

space, but we shall not take up this question until V.5, in connection with an
arbitrary Teichmller space.
In section 5 we introduce the inner radius of univalence of a simply connected domain. Its use makes it possible to analyze further the image of the
universal Teichmller space in the space of Schwarzian derivatives.
1. Models of the Universal Teichmller Space

1.1. Equivalent Quasiconformal Mappings
Let us consider the family of all quasiconformal mappings of a fixed domain
in the plane. In this section we assume that this domain is the upper halfplane H. We wish to introduce additional structure to this family and begin
by regarding two mappings as equivalent if they differ by a conformal mapping. In view of the Riemann mapping theorem, we may then restrict ourselves to self-mappings of H and require that they are normalized so as to
keep fixed the three boundary points 0, 1 and
We denote by F the family
of such normalized mappings. (Recall: every element' of F can be extended to
a homeomorphic self-mapping of the closure of H. It is actually the extended
mappings to which the normalization requirements apply.)
By the existence and uniqueness theorems for Beltrami'equations (Theorems 1.4.4 and 1.4.2), there is a oneone correspondence between F and the
on H.
open unit ball B of the Banach space which consists of all
A more interesting space is obtained if we introduce a weaker equivalence
relation.
Defnition. Two mappings of the family F are equivalent if they agree on the
real axis. The complex dilatations of equivalent mappings are also said to be
equivalent. The set of the equivalence classes is the universal Teichmller
space
We thus have two models for T: Its points are classes of equivalent mappings i'i the family F or of equivalent functions on the ball B.
A third model is obtained in terms of quasisymmetric functions. We recall
that a quasisymmetric function is said to be normalized itit fixes the points 0
and 1. Let X denote the class of all normalized quasisymmetric functions. If
III. Universal TeicbmUller Space
98
[f J is the point of T represented by the mapping fe F, then

(1.1)
is a bijective mapping of T onto X. For it is clear from the definition of T that

(1.1) is well defined and injective. By Theorem 1.5.1, it maps T into X, and by
Theorem 1.5.2, it is surjective. It follows that we can rephrase thedefinition of
the universal Teichinller space:T is the set of all normalized quasisymmetric
functions.
This observation allows an important conclusion:
Theorem 1.1. Every point of the universal Teichmller space can be represented
by a real analytic quasiconformal mapping f EF or by a real analytic complex
dilatation p e B.
PROOF. The result follows immediately from Theorem 1.5.3. (For a complete
proof, see 11.5.2.)
We shall see in V.3.2 that the universal Teichniller space contains as a

subset the Teichmller space of any Riemann surface which allows a halfplane as its universal covering surface. It was Bers [7,8] who recognized the
importance of this largest and, in many way's, simplest Teichmller space and
gave it the name universal.
J.2.
1ff belongs to F, then so does its inverse f'; along with f and g in F, the
compositionfog is also in F. The family F can thus be regarded as a group.
From the definition of the universal Teichmller space it follows that T
inherits this group structure: T is the quotient of the group F of all normalized
quasiconformal self-mappings of the upper half-plane by the normal subgroup of
mappings equivalent to the identity.
1ff g e F, the rule
(f]o[g]= [fog]
(1.2)
defmes the group operation in T The neutral element, i.e., the point of T
determined by the identity mapping (or by the complex dilatation which is
identically zero) is called the origin of T and denoted by 0.
Normalized quasisymmetric functions also form a group under composition. This follows from Theorems 1.5.1 and 1.5.2 or from an easy elementary
computation. We see that the mapping (1.1) is an isomorphism between the
groups T and X.
Let us consider,
a moment, quasiconformal self-mappings of II which
are not necessarily normalized. If 11 and 12 are two such mappings, we still
say that
is
to f21f
is the identity. Let JEF be a
99
1. Models of the Universal Teichmlier Space
normalized and g an arbitrary quasiconformal self-mapping of H. We choose

eF,
a Mbius transformation h, mapping H onto itself, such that
and defme w = w151: T , T by the formula
co([f]) =
Then w, which depends only on the equivalence class [g], is a well

defined bijection of T onto itself. Further, [g] = 0 implies that w is the
identity and
[h1
co[g,]01921([f]).
We conclude that when g runs through all quasiconformal self-mappings of
H, the transformations
T ,
T form a group. It is called the universal
modular group M.
We now return to normalized quasiconformal self-mappings of H and

consider the subgroup M, of the universal modular group consisting of transformations
with g e F. Then
=
are right translations of the group T
elements of
In 2.1 we shall introduce a metric into T and prove that every
i.e.,
is an
isometry with respect to this metric.

The group M, of right translations is transitive: If
there is an
of1..
clearly the case if g =
are given points of
[fr] and P2 = [f2]

This is
such that P2 =
1.3. Normalized Conformal Mappings

The mappings belonging to F can be continued quasiconformally to the
plane by reflection in the real axis. However, such an extension does not give
new insight into the properties of T It was a fundamentai observation of Bers
[4] that one should extend, not the mappings of F but rather their complex
dilatations, in such a way that the corresponding extended mappings are

conformal in the lower half-plane. The machinery developed in Chapter II
can then be applied to the study of T
Let
B and f'4 be the mapping of F withcomplex dilatation i. We extend
,u to the lower half-plane H' by giving it' there the value 0. Let
be the
quasiconforinal mapping of the plane which 5*ce 0,1, co and whose complex
is conformal.
dilatation agrees with the extended Then
Theorem 1.2. The complex dilatations and v are equivalent

the conformal mappings f,LI H' and H' coincide.
and only
and
PROOF. Suppose first that
H' f,IH'. The mappings j a
are both conformal in the upper half-plane H, which they map
III.
100
TeichmSller Space
onto the same quasidisc. Because they fix 0, 1, oo,it follows that they agree in
H, and hence also on the real axis R. Since f,, = f, on R, we condude that
= on R, i.e., p and v are equivalent.
Assume, conversely, that f" =1' on R. We define a mapping w of

in f,(H' '.i R), and w =
the plane by the requirements w =
of,1
in
o
of'
f' on R it follows
the
hypothesis
From
f,(H).
f,,
that w is a homeomorphisin of the plane. In addition,
But so is also wjf,(H), because f,,o(fT'
is conformaL
and f'of,1 arc conformal. Since
f,(R) is a quasicircie, we infer from Lemma 1.6.1 that w is a Mbius transformation. Owing to the normalization, w is the identity mapping and so
Let F* be the family of all quasiconformal mappings of the plane which fix
the points 0,1, and arc conformal in the lower half-plane. Two mappings
f,, and f, of F* are said to be equivalent if they agree in the lower half-plane.
Theorem 1.1 says that every equivalence class [f"] contains real analytic
mappings. It follows that each class [1,3 has representatives which arc real
analytic in the upper half-plane H. For in H,
where f,1o(fT' is conformal. Therefore, f,,IH is real analytic
is.
In particular, there arc mappings f,, which are conformal in H' and real
analytic in H but which, nonetheless, are very irregular on the real axis R.
ban be arbiWe recall that the Hausdorif dimension of the image cur e
trarily close to 2 (cf. 1.6.1).
By Theorem 1.2 the space T can be regarded as the set of the equivalence
classes [1,3. Or more explicitly: The universal Teichmller space is the set of
the normalized conformal mappings
1.4. Sewing Problem

The characterization of T by means of the conformal mappings
leads
to far-reaching conclusions. Section 4, in particular, will be devoted to considerations emanating from this model of
Anticipating a need in section 1.5, we give a solution to the following
sewing problem: Let h be a strictly increasing continuous function on the
real axis, growing from to + cc. Find conformal mappings 11 and f2 of
the upper and lower half-plane, respectively, onto complementary Jordan
domains such that
fr1ofz=h
normalized if and 12 both fix 0, 1
on the real axis. We call the pair
and cc.
Depending on h, the sewing problem, to which many questions in complex
1. Models o( the Univeiial Teichmikr Space
101
analysis seem to lead, need not have any solution or it may have infinitely
many pairs of solutions. However, if h is a normalized quasisymmetric function, the existence of a unique nonnalized solution can be easily established
It is in this form that the
by aid of the quasiconformal mappings f and
result will be used for studying the universal Teichmlier space.
Lemma 1.1. Let h be a normalized quasisymmtric function. Then the sewing
problem has a unique normalized pair of solutions.
PROOF. Given a function he X, there is a mapping f' e F such that f"I ft = h.

Then
is a solution of the sewing problem. This can be verified immediately.

Suppose that the pair
is also a normalized solution. Then g2IR
oh
ofUIR. Hence, the mapping w which agrees with of's in H u ft
and with g2 in H' is a homeomorphism of the plane. OfT the real axis it is
quasiconformal. By Lemma 1.6.1, w is quasiconformal everywhere. Since w
has the same complex dilatation as and both mappings fix 0, 1 and CX), it
follows from the uniqueness theorem (Theorem 1.4.2) that w =
Comparison of the definitions of w, and 12 then shows that g1 = f1, g2 = 12.
0
Note that 11 and f2 map the
onto quasidiscs. Lemma 1.1 is due
to Pfluger [21; in [LV], p. 92, it was proved without the use of the existence
theorem for Beltrami equations.
1.5. Normalized Quasidiscs

We shall now express in geometric terms the fact that points of the universal
We
TeichmUller space can be represented by the conformal mappings
call a quasidisc normalized if its boundary passes through the points 0, 1,
and is so oriented that the direction from 0 to 1 to IX to 0 is negative with
respect to the domain. Let A denote the class of all normalized quasidiscs.
1ff E Fe, then
[f]*f(H')
(1.3)
is a bijective mapping of Tonto We first conclude from Theorem 1.2 that

of is a conformal self-mapping
(1.3) is well defined. If f(H') = g(H'), then
Hence
of H' fixing 0, 1,
fIH' gIH', i.e., [f) = [g], and it follows that
(1.3) is
Finally, by Lemma 1.6.2, every quasidisc is the image of H'
under a quasiconformal mapping of the plane which is conformal in H'. Since
the required normalization is achieved by use of a suitable Mbius transformation, we conclude that (1.3) is suijective.
IlL Universal
The bijection (1.3) provides one more model for the universal Teichmfler
space: Tis the collection of all normalized quasidiscs.
Using Lemma LI we obtain a connection between nonnalized quasidiscs

and the group structure of T(Gardiner [1]).
Theorem 13. Tho points [f"], [f"] Tare inverse elements of the group Tjf
and only
real axis.
and f,(II') are mirror images with respect to the
the
PROOF. Assume first that

and [fVJ are inverse; we can then take
be the quasiconformal mapping of the plane which fixes
= (f'Y'. Let
vanishes in H and
and whose complex dilatation
the points 0, 1,
=
and denote by g2
at almost all points zeH'. We write 9i
equals
mapping of H' onto f,4.(H') which keeps 0, 1,c fixed.
the unique
Then 9i and 92 are normalized conformal mappings of the upper and lower
half-planes, respectively, onto complementary quasidiscs.
og2 on the real axis R, we continue fP by reflection
In order to study
in and use the same notation f" for the extended mapping. Then
in H', because both sides are normalized quasiconformal self-mappings of H'
with the same complex dilatation. Hence, on R

og2
= (fMYl = 7.
Now set
Ii =
f1 =
and f2 are also normalized conformal mappings of the upper

and lower half-planes onto complementary quasidiscs. On the real axis,
of2 = 7 =
og2. We conclude from Lemma 1.1 that g1 =
g2 = 12.
one way to verilS'
it follows that
=
From the definition of
derivatives. Since f,(H) = f1(H) = g1(H) =
this is to compute thr
Then
we obtain
f,(H') =
and the first part of the theorem has been proved.

After this the converse is easily established. Suppose that f,6(H) f,(H').
where is
By what was just proved, there is a quasiconformal mapping
determined by fA = (fm', such that
= fA(H'). Since the mapping (1.3)
is injective, we conclude that A is equivalent to v. It follows that ff") and
[7) are inverse elements of
0
In 11.2.1 we defined the distance (f(H')) =
i),,. between the domains
f(h") and H'. This notion was generalized in 11.2.7 to apply to two arbitrary
domains cotiformally equivalent to discs. When the domains are normalized
103
2. Metric of the Universal Teichmller Space
quasidiscs, we can give a new definition: if J,

tween f(H') arid g(H') is defined by
q(f(H'), q(H')) =
J.
then
'I
By use of the
It is easy to cheek that q, so defined, is a metric on
bijection (1.3), the metric can be transferred to T This q-metric will be
in detail in section 4. Before that in section 2, a different metric will be defined
for Tin a more direct manner.
2. Metric of the Universal Teichmiler Space

2.!. Definition of the Teichmller Distance
In addition to the group structure, the universal Teichmller space has a
natural metric. We obtain this metric by measuring the distance between
quasiconformal mappings in terms of their maximal dilatanons. When representing points of T mappings fit does not matter whether we assume that
fe F (normalized se
of H) or that J e P (normalized mapping of
the plane. confotmal in H'). This follows from the fact that
and f,, have the
same maximal dilatation.
The distance between the points p and q of T defined by
r(p,q)=
where K denotes the maximal dilatation and
is
or f
called the Teichmller distwice between p and q. Teichmller [1] used this
idea for defining distance in his studies on compact Riemann surfaces
V.2.2).
Before proving that the Teichmller distance makes T into a metric space.
we show that r admits various other formulations. In order to fix the ideas.
we assume for a moment U at the
mappings representing
points of T are in the class F*.
Fix a representative
p and set
(p,q) =
Kqoj6tlgeq}.
(2.2)
Alternatively, we can fix both e p and 9o eq. consider the class W of all
quasiconformal mappings of the plane which agree in f0(H') with
and set
r2(p,q) =
W}.
(2.3)
Lemma 2.1. The functions r,
PROOF.
Clearly, z
and r2 are the same.
If we W, then g = wof0eq, so that r1 r2. Finally,

-
III. Universal Tcicltmiiller Sp.cc
104
By
our previous remark, Lemma 2.1 holds also if the points of T arc
represented by mappings from F. The class W must of course be modified,

self-mappings of H which agree
being now the family of all
we thus reencounter the family we
on R with g0
If we set h = g0
considered 1.5.7, there under the name F,,. We proved that F,, contains an
extremal mapping which has the smallest maximal dilatation in F,,. If this is
denoted by K,,, then by Lemma 2.1,
.
(2.4).
One important consequence of the existence of an extremal mapping in F,,

is that in expressions (2.1), (2.2), and (2.3), we can replace iilf by mm. After this
observation, it is readily seen that (2.1) (or (2.2), (2.3), or (2.4)) defines a metric
in T. Clearly v is non-negative and symmetric, and r(p,p) =0. Suppose that

r(p, q) = 0. Then it follows from (2.4) that F,, contains a conformal mapping
and because of the normalization, this mapping is the identity. Hence f0IR =
which implies p = q. The triangle inequality follows from the property
K,01 K,K, of the maximal dilatation.
Th6 Teichmller distance is invariant under the universal modular group. For
the subgroup M,, the trivial obseivation
9of1 =
yields immediately the invariance
r([f], [j]) =
r([f of0 1], [go
In particular,
r([f],[gJ) =
so
that all distances can be measured from the origin. The general result
(which we shall not make use of) follows from the fact that conformal mappings do not change maximal dilatation.
2.2. TeichmUller Distance and Complex Dilatation

If f and g have the complex dilatations and v, the norm of the complex
dilatation of g of -' is equal to
v)/(1 giv)
Therefore, in terms of
Q
complex dilatations, the Teichmilller distance assumes the form
.
og
mink
IRii
v)/(1
1 IIOz v)/(1
lI4ep,veq
The right-hand expression displays a striking similarity to the hyperbolic

distance h in the unit disc D given by formula (1.1) in 1.1.!. We introduce the
number
II
v) fl
= ess sup h(js(z), v(z))

ZGH
and call it the hyperbolic distance between the complex dilatations p and v.
105
2. Metric of the Universal TeichmUller Space
From (1.1) in 1.1.1 we then obtain the characterization

r(p,q) =
of the Teichmller distance as the minimal hyperbolic distance.

Consider the bounded function
8(p,q)=inf[ LV
pep,veq
I.
From fi = tanh r we see that in the definition of fi, inf can be replaced by mm,
that (I also makes T into a metric space, that is invariant under the universal
modular group and that the metrics defined by r and are topologically
equivalent. In other w.irds, in studying topological properties of Twe may
use the metric provided by instead of the TeichmUller metric. Let us give a
simple application.
Theorem 2.1. The universal TeichmuUer space is pat hwise connected.
PROOF. Consider
the origin of T, i.e., the point represented by the function of
B which is identically zero, and an arbitrary point p e 1' represented by

1, let p, be the point represented by the function tp of B. Then
0
We see that the mapping t
p,
For
is continuous, i.e., it is a path in T joining
the origin to p.
In 3.3 we shall prove that the universal Teichmller space is not only
pathwise connected but even contractible. This means that T can be deformed continuously to a point. The result is less trivial than it would seem
at first glance, because for equivalent and v, the complex dilatations tj,t and
tv need not be equivalent for 0 < t < I (Gehring [1]).
2.3. Geodesics for the Teichmller Metric

r) is the supremum of
y(t1)) for
= 1 of the unit interval. An arc y is a
all subdivisions 0 = t0 <t1
geodesic if the length of every subarc of y is equal to the distance between
the endpoints of
Geodesics of T can be described explicitly with the help of extremal complex dilatations. We say that jz e p is extremal if If f( = mm { fi v (V e p}.
The length of an arc y: [0, 1] *
Theorem 2.2.
(7',
is an extremal complex dilat ation for the point p e T, then
'
(.25
HI. Universal Teichmller Space
106
is
extremal for the
p1
(j.t,j. The arc t
is a geodesic from 0 to p, and

(2.6)
PROOF. From (2.5) we see that p,(r) is the point which divk!cs the hyperbolic
in the ratio t: (1 t)
from 0 to
length (in the unit disc) of the line
(cf. formula (4.16) in 1.4.7).
has maximal dilatation K, then by Theorem 1.4.7, the mapping
If
o
has maximal dilatation K1 '.
maximal dilatation Kt and
owEp, and so K c K,
that wept. Then q' =
Consequently, K,, K'. We conclude that ji, is extremal for the point
0), i.e., the validity
This reasoning also shows that
0) 4Iog K' =
has
Suppose
of (2.6).
has maximal dilatation K'', we conclude that z(p,,p)

=
for every t. Finally, if
r(O,p,) +
we repeat the
see that
arguiiient for an arbitrary subarc of t
t'p,isageodesic.
.0
If
(1
t)t(p,0).
Since the extremal i need not be unique (cf. 1.5.7). we cannot conclude that
the geodesic e ' ft,] is unique.
of the Universal TeichmUller Space
2.4.
We shall now prove that the space (T, is complete. We first describe explicitly the construction on which our proof of the completeness is based and
list the pertinent facts associated with that construction.
Lemma 2.2. Every Cauchy sequence in (T, i) contains a subsequence whose
points are represented by complex dilatations with the following properties;
10 urn
20 [fr]
3
40
.f."(z)
exists almost everywhere;

in the Teichmller metric;
uniformly in the spherical metric;
locally uniformly in the upper half-plane.
order to simplify the notation, we renumber functions each time

write =
pass from a sequence to its subsequence.
be a Cauchy sequence in (T, t). We shall construct inductively, a
Let
properties 0_40 using suitably chosen mappings
with
First, fix a mapping f1 so that
PROOF. In
inf log
for each p. the infimum is taken over all mappings of [J4,,]. Since
is a Cauchy sequence, such a mapping f1 exists, as can be seen from
formula (2.2) and Lemma 2.1. We renumber the sequence by setting f1 = f1.
where
Metric of the Universal Teichmller Space
After this, we choose for every n >
107
the mapping f0 from its equivalence
class so that
log K10011,
From this new sequence (fe) we choose a mapping
where
so that
again for each p the infimum is taken over all mappings of the class
We set = f2, and for n> 2, choose a representative of [fe] so that
logK1012,
Continuing this procedure we obtain a sequence (f0), such that ([fe]) is a
subsequence of the given Cauchy sequence and such that, for any two consecutive indexes, the maximal dilatations satisfy the inequality
log
of,;' <Z2",
log K1,
of'
ii
= 1, 2
It follows that
2-(Jl+J1)
(2.7)
for n, p = 1, 2
Considering the connection between the maximal dilatation and the norm
of the complex dilatation, we deduce from (2.7) that the complex dilatations
ic,, off0 satisfy the inequality
Pn+p
Lanfl
Thus (p,,) is a Cauchy sequence in

Since
is complete, the limit
p = urn p,, exists in
Thus the validity of condition 1 follows. From (2.7)
we conclude that the mappings f0 (and hence also
are K-quasiconformal
for a fixed K. It follows that
< 1. Therefore, [p] = limfp0
(T, fi) and
hence also in (T, t). This means that the statement
is true. Since the
mappings f,, and
keep the three points 0, 1, oc fixed, the families { f0) and
{ f"} are'normal, by Theorem 1.2.1. Hence 3 and 4 follow, after possible
passage to further subsequences.
D
Theorem 2.3. The universal Tei'chmller space is complete.
PRooF. In view of statement 2 in Lemma 2.2, it is enough to observe that if

a Cauchy sequence contains a convergent subsequence, then the sequence
itself is convergent.
El
Having proved Theorem 2.3 we see that in Lemma 2.2, we need not pick a
subsequence: If p0 . p in (T, r), there are complex dilatations e p0, pep, such
that the conditions 1, 3, and 4 of Lemma 2.2 hold (cf. Theorem 1.4.6).
III. Universal Teichmller Space
108
In view of the simple relation between the distances t and fi, it is easy to
show that the metric space (T, is also complete. In contrast, using Schwarzian derivatives we shall obtain in section 4 one more model for T which is a
metric space homeomorphic to (T, r), but is not complete. (It is, in fact, the
metric space (A,q) discussed at the end of 1.5.)
3. Space of Quasisymmetric Functions

3.1. Distance between Quasisymmetric Functions
Let us consider again the space X of normalized quasisymmetric functions
discussed in 1.1. For heX we defined in 1.5.1 the maximal dilatation
Imitating the method we used in defining the Teichmller distance, we set
p(h1,h2) =
for h1, h2 cX. It is easy to verify that p defines a metric on X.
Theorem 3.1. The group isomorphism
[f]-'fIR
(3.1)
is a homeomorphism of(7', v) onto (X, p).
We proved in 1.1 that (3.1) is a bijection of T onto X. From (2.4) and

the left-hand inequality (5.10) in 1.5.7 it follows that
PROOF.
R,f2
R)
1 [121).
(3.2)
Hence (3.1) is continuous. From Lemma 1.5.5 (or from the right-hand inequality (5.10) in 1.5.7) we conclude that the inverse of(3.1) is continuous.
D
From the double inequality (5.10) in 1.5.7 we can draw another conclusion:
The space (X, p) is complete. For we conclude from the right-hand inequality
(5.10) that the preimage of a Cauchy sequence in (X,p) is a Cauchy sequence
in (T, r). The inequality (3.2) then shows that (3.1) maps a convergent sequence of (T, r) onto a convergent sequence of (X, p).
Suppose that h, hNEX, n = 1,2,..., and that
= 0. Then
h
locally uniformly in the euclidean metric. For by Lemma 1.5.1,

is a
family. If is the limit of a convergent subsequence
of (he), we
= 1. Hence = h. Since every convergent subhave Kt01,-, lim
limit Ii, the sequence itself tends to h.
The converse is not true. A counterexample is obtained if we set
=x
for x n, and h(x) = 2x n for x > n. Then
X and urn
h(x) = x,
uniformly on every bounded interval. But the quasisymmetry constant of
is 2, so that by the remark in 1.5.2, urn inf
It) > (log ).' (2))/2 > 0.
sequence of (ha) has
109
3. Space of Quasisyinmetric Functions
3.2. Existence of a Section

Using the space (X, p), we can prove that the universal Teichmiiller space is
contractible, i.e., there exists a continuous mapping of T x {tjO t 1} into
T which is the identity mapping for t = 0 and a constant mapping for t = 1.
As a preparation for the proof, we modify the mapping (3.1) by changing
its domain of definition. Let

be the space of functions bounded and
measurable in the upper half-plane, and B {p e
<1 } its open
unit ball. For pe B, we now consider the mapping i/i of B onto X, defined by
=
In view of the definition of the Teichmuller distance, inequality (3.2) can be
written in the form
p (fiR
IVIR)
v)/(l
(p v)/( 1 gv)
It follows that .,1i is continuous.

Of course, the mapping B ' X is not invertible. However, there exists a
section s: X ' B, i.e., a continuous mapping of X into B such that os is the
identity mapping of X.
A section s can be constructed with the aid of the BeurhngAhlfors extension of a quasisynunetric function. Given a function he X, we set as in 1.5.3,
f(x+iy)=1
Let
Jo
(h(x+ty)+h(xty))dt+iI (h(x+ty)h(xty))dt.
Jo
be the complex dilatation of f. We shall prove that the mapping
defined by s(h) = p,is a section.

X, 1
1, 2, we denote their BeurlingAhifors extensions by and
set s(h1) =
Let K be the maximal dilatation and k the quasisymmetry
constant of h2 o
We showed in 1.5.2 that k ).(K). Hence,
s: X * B,
For
p(h1,h2) =
>
Now suppose that h2 h, in (X,p). It follows from the above inequality

that k I. By Lemma 1.5.3, the maximal dilatation of .12
then tends to
1. This is equivalent to P2 converging to Mi, and we have proved that s is
continuous.
Trivially, (i4' os)(h)
= h, so that os is the identity mapping of X.
Consequently,s: X -.. B is a section.
3.3. Contractibility of the Universal Teichm tiller Space

After these preparations, the desired result can be easily established (Earle
and Eells [1]).
Theorem 3.2. The universal Teichmller space is contractible.
III. Universal Teichmuller Space
110
PRooF. Every point of T is an equivalence class [s(h)], h X. We show that

([s(h)),t)*[(1 t)s(h)J
(3.3)
deforms T continuously to the point 0 as t increases from 0 to 1.
In proving this, we make use of Theorem 3.1 which says that

is a homeomorphism of (T, 'r) onto (X, p). It means that instead of (3.3), we
can consider the induced mapping
(h,
(3.4)
t)
of X x [0, iJ into X. Clearly, (h,0) ' h and (h,

follows if we prove that (3.4) is continuous.
1)
identity. The theorem
The mapping (3.4) is the composition of the three mappings

(h, t) . (s(h), t),
(s(h), t) (1
t)s(h),
(1
t)s(h)
The first one is continuous, because we just proved that h ' s(h) is a continuous map of X into B. The second maps B x [0, 1) continuously into B,
t1)s(h1) (1
Final
+ It1
maps
the third mapping is continuous, because we showed that p
B continuously into X. Hence, (3.4) is a continuous contraction of X to a
point, and (3.3) has the same property with respect to T
D
since
ly,
3.4. Incompatibility of the Group Structure with the Metric

We showed at the end of 3.1 that pointwise convergence of functions of X
does not imply p-convergence. A slightly more complicated counterexample
leads to the conclusion that the topological structure and the group structure
of X are not compatible. We express the result in terms of T.
Theorem 3.3. The universal Teichmller space is not a topological
PROOF, The
theorem follows if we find an
[f]
group.
T and a sequence of points
that [ga] tends to [g] but [fog,,] does not tend to [fog].
Because the mapping (3.1) is a group isomorphism and a homeomorphism,
the counterexample can be constructed in X. We follow a suggestion of
such
P. Tukia.
In order to simplify notation we write instead of fj IR. We define f

as follows: f(x) = x if x 0. f(x) = x/2 if 2 x <0, and f(x) = x + I
if x < 2. Then f is a 2-quasisymmetric function of X. Set i,,(x) = x
Then
is
x > 0 and z,,(x) = (1 + 1/n)x if x <0, n = 1, 2
if
(I + l/n)-quasisymmetric, and therefore (1 + 1/n)2-quasiconformal (cf. 1.5.3).

If denotes the identity mapping of onto itself, we thus have
i
log(1 + 1/n).
Let us define g,, =
Because
we deduce that
Ill
-1, =
(3.5)
of -1 (which converges to the identity i pointf

= un the p-metric.
wise) does not tend to Jo (urn
We prove that f o
Direct calculation yields
It follows that
+ lfn)x
+ ljn)x +
if
I
n,'(n 4-
x <0,
if 1 x < n/(n +
1).
has a quasisyrnmetry constant 2 for every n.
Consequently,
in conjunction with (3.5). this shows that (X, p). and hence (T, t). is not a
:opological group.
D
translation, the left translation [f]
We sce that, unlike the
iixed, aced not be continuous in T.
{f0 of],

4.1. Mapping into the Space of
Derivatives
The universal Teichmller space was defined by means of quasiconformal

self-mappings of the upper half-plane. In this scction and in section 5, we
Now
is a self-mapping
the roles of the upper
lower
of the lower half-plane H' and f,, is conformal in the upper half-plane H. This
change, which does not frect any of the results in sections 13, simplifies
notation here, because we are now dealing primarily with the conformal part
of the mappings
It follows from what we proved in 1.3 that each point of the universal
Teichmller space T can be represented by a normalized conformal mapping
It was Bers [6] who noticed the importance of forming the Schwarzian
derivative and defining the mapping
[p]'SIIH
of T The image points, as Schwarzian derivatives, are holomorphic functions
in H, for which the norm defined in 11.1.3 is pertinent.
This leads us to introduce the space Q of all functions o holomorphic in H
for which the hyperbolic sup norm
II
Ii = sup4y2lp(z)t,
zH
III. Universal Teichmtiller Space
112
= x + ly, is finite. The space Q has a natural linear structure over the
complex numbers.
Furthermore, Q is complete. For if

is a Cauchy sequence in Q, then
is a Cauthy sequence in
is complete, there
Since
=
is a
such that
converges to in
Here can be taken to be
continuous. Then q,, converges locally uniformly to q =
because is
locally bounded. It follows that p Il holomorphic,
, 0, and p Q.
We conclude that Q is a Banach space. Its points are Schwarzian derivatives:
By Theorem 11.1.1 every function p e Q is the Schwarzian derivative of a
function f meromorphic in H.
By Theorem 11.1.3,
Going back to the functions f,,, we write sp =
H
6. Therefore, [pJ ' maps T into Q. The mapping is well defined, for
if v is equivalent to we have
and hence s,, =
=
II
ii
4.2. Comparison of Distances

We shall prove that the mapping [p3 -+ is a homeomorphism of T onto its
image in Q. To this end we shall compare the a-distance of two given points
[p] and [v] of T to the distance of their images s,, and in Q. We write
(P2) = (p1 (I'I II for points of Q.
II
In the special case v = 0, estimates in both directions can be obtained

directly from our previous results. If = 0, then also = 0, and q(s,1, 0) =
By Theorem 11.3.2,
This
ii. Moreover,
=
holds no matter how p is chosen from the equivalence class. Consequently,
6fl([p],0).
(4.1)
is equal to the distance

We remark that
of
from H (cf.
the remark at the end of 1.5).
In order to get an inequality in the opposite,, direction, we choose an
arbitrary eQ such that II ii <2. By Theorems 11.1.1 and 11.5.1, there is a
normalized quasiconformal mapping f of the plane which is conformal in H,
for which S1 1K = (P,and whose complex dilatation p in the lower half-plane
is obtained from the formula
= 2y2 q,(z), z H. For this mapping f =
we have ii
=
11/2. Hence,
(4.2)
We assumed that
0) < 2. But since all a-distances are <1, (4.2) holds
trivially if
2.
We shall now generalize (4.1) and (4.2) for arbitrary points [p] and [v]. We
start with the transformation rule

liSp SVIIH
= IS1
for the Schwarzian derivatives (formula (1.10) in 11.1.3). We write

and apply Theorem 11.3.2 to the conformal mapping w =
(4.3)
= f,(H)
in the
4.
of
113
Derivatives
quasidisc .4,. It follows that
oo(AJII
here o0(A,) = 6 + (,,) is the outer radius of univalence of A,. In view of

(4.3), this yields the estimate
[v]),
(4.4)
which is the desired generalization of(4.1). Since a0(A,) .12, we could use as
the coefficient on the right-hand side the absolute constant 12. On the other
hand, a0(A,) can be replaced by
where AM fP(").
It is more difficult to generalize the inequality
We choose v from its
equivalence class so that f, has the smallest possible maximal dilatation K,..
After this, we consider Schwarzian derivatives which are so close to s, that
s is the constant of Theorem 11.4.1. Then, by formula
(4.3),
HSWIIA.
<s(K,).
We know that w has a quasiconformal extension, namely, f_of,'. However,

we prefer to extend w by utilizing Theorem 11.4.1, which makes it possible to
estimate the complex dilatation. By that theorem, w has a quasiconformal

extension to the plane such that the complex dilatation K of the extended
mapping satisfies the inequality
(K,,)
45
If the extended w is also denoted by w, then = w of, is a quasiconformal

to the lower half-plane. Thus A. is equivalent to p. Because
extension of
o
',we
have,
therefore,
w=
=
fJ([p],[vJ).
Combining this with (4.5) we finally arrive at the inequality
e(K,)/3({uJ,[v]),
(4.6)
valid also if
e(K,). This contains (4.2) as a special case: If v = 0,
then K, = I and E(K,) = 2. Since the roles of p and v can be interchanged, we
e(K,fl.
can replace
in (4.6) by
4.3. Imbedding of the Universal Teichmller Space

The estimates (4.4) and (4.6) show that the and q-metrics are topologically
equivalent. Thus a new important model is obtained for the topological space
(T,t).
14
Theorem 4.1. The mapping
[/4]
a homeomorphism of the universal Teichinller space onto its image in Q.
We noted already in
that (47) is well defined in 7 If [p3 and [v]
ave the same image, it follows from the normalization that
= f,IH, i.e.,
and v are equivalent. Hence (4.7) is injective. Inequality (4.4) shows that
i.7) is continuous, and (4.6) that its inverse is continuous.
0
In 4.5 we shall show that the image of T under the homeomorphism (4.7)
open in Q. (We have almost proved this in establishing (4.6).) The mapping
which will later be considered in connection with an arbitrary Teichnller space, is called the Bers imbedding of TeichmUller space.
' [it] in
By Theorem 4.1, the convergence
Sfl,. in Q implies that
*
Hence, by Lemma 2.2 and the remark following Theorem 2.3,
uniformly in H.
Anticipating developments in Chapter V, we denote the image of T under
(4.7) by T(1). When there is no fear of confusion, we often identify T(1) with
universal Teichmller space. Like X, the space T(1) is simpler than T in
that its points are functions and not equivalence classes of functions.
We can also define T(l) = {Sf(f is conformal in H and has a quasiconformal extension to the plane}. For such an f is equal to a normalized
mapping
modulo a Mbius transformation, which does not change the
Schwarzian derivative.
By Theorem 11.5.1, the set T(l) contains the open ball B(O, 2)
<2). In this ball, the inverse of the mapping (4.7) can be

described explicitly:
[p],
2y2q(z).
The space (T(l), q) is not complete, even though it is homeomorphic to the

complete spaces (T,r), (T,fl) and (X,p). In order to prove this, it is sufficient
to find an Sf E Q\T(1) and functions S1e T(1), n = 1, 2, ..., such that Sf ' S1
in Q. Then (S1) is a Cauchy sequence in T(l) but its limit is not in T(l). An
=
example is provided by the functions z -+ f(z) = logz, z
in H,
which we considered, for another purpose, in 11.1.4. Since
S1,,(z) = (1
= 2(1 1/n2) < 2. By Theorem 11.5.1, the

has a
quasiconformal extension to the plane. Hence S1e T(1). In 11.1.4 we saw
already that
we have
S,Ij = 2/n2 -+0.

But since z * log z does not even have a homeomorphic extension, Sf is not
in T(l).
115
4. Space at Schwanian Derivatives
4.4. Schwarzian Derivatives of Univalent Functions

Let us define the set
u=
univalent in H).
Then trivially T(1) U, and by Theorem 11.1.3, U c Q. More precisely, we

conclude from Theorem IL 1.3 that U is contained in the closure of the ball
B(O, 6)
{4 Q 114' II <6}. On the other hand, it follows from Theorem 11.5.2
that U contains the closure of B(O, 2).
Let f1(w) = w + 1/w,f2(w) = w l/w. If h is a conformal mapping of the
upper half-plane onto {wlfwI> 1), then S1101,, 5,20,16 U. From the calculations in 11.2.6 it follows that
S120,. II = 12. We conclude that the
diameter of U is 12.
The set U is closed in Q. For suppose that
eU, n = 1, 2, ..., and that S,
converges to S1 in Q. We show that f is univalent.
We are free to compose the functions with arbitrary Mbius transformations. There is no loss of generality, therefore, in assuming that every f,, fixes
the same three points a1, a2, a3 in H. By Theorem 1.2.1, the family
is then
which is locally uniformly
normal. Consequently, (fe) contains a
convergent in H. By renumbering the functions we may assume that (fe) itself
I
has this property. The limit g =
fixes a1, 02 and a3, and is therefore univalent in H. At every point z E H we have urn
= S9(z) and also
urn S1(z) = Sf(z). Hence, I differs from g by a Mbius transformation, and so
f is univalent.
Since U is closed and T(l) U, the closure of T(1) is contained in U. If
U, we can always find functions with
T(1) such that
f(z) =
locally uniformly in H.
An approximating sequence
z
with
+ i/n
H onto the disc
T(I) is obtained as follows. Set
n=
(4.8)
= {wlIw
2, 3
iI <((n 1)/(n
+ 1))Iw + it),
exhaust H, and g,,(z) -+
locally uniformly. Hence, by setting J, = fog,,, we obtain a sequence of
functions for which (4.8) is true. The property
T(1) follows from the fact
that f,,(R) = f(aD,,) is a quasicircie (cf. the remark in 1.6.1).
If (4.8) holds, the derivatives off, converge to the derivatives off. Hence,
Then
maps
whose closure lies in H. As n *
the discs
S1(z) = urn S1(z)

fl-4
locally uniformly in H. However, as we showed in 11.1.4, it does not necessarily follow that
in Q. We cannot conclude, therefore, that the
'
closure of T(1) coincides with U. A counterexample such as the one in 11.1.4
does not disprove this either, but actually the closure of T(1) is not the whole
of U. This will be explained in 4.6.
III. Universal TeichmUller Space
116
4.5. Univalent Functions and the Universal TeichmUller Space

From Theorems 11.4.1 and 11.4.2 we obtain a remarkable connectibn between
the sets T(l) and U (Gehring [2]).

Theorem 4.2. The set T( 1) is the interior of U.
PROOF. We prove first that T(1) is an open subset of Q. Fix an arbitrary point
S1 of T(l). For ShEQ we write g = hof1, and conclude that g is meromorphic in the quasidisc f(H). By Theorem 11.4.1, there exists a positive constant
e such that if II IIf(H) <e, then g is univalent in f(II) and has a quasiconformal extension to the plane. Now choose She Q such that II Sh S, HR <c.
Then
IISgIIj1>
ItS,, SI liii
<
h = gof, we conclude that S,,E T(1). It follows that T(l) is pen.

Since T(1) U, the proof will be complete if we show that mt U c T(1).
Choose a point
mt U. We then have an > 0 such that
Because
Let g be an arbitrary meromorphic function in the domain f(H), with the

property S5 II ((H)
If h = g of, then
= HSSIIJ(u) 8.
It follows that S,,e V c U, i.e., h is univalent in H. But then g = hof' is
univalent in f(H). What we have proved is that f(H) is an e-Schwarzian
domain. Hence, by Theorem 11.4.2, the domain f(H) is a quasidisc. We
conclude that S1e T(l) (cf. Lemma 1.6.2, statement 30) as we wished to
show.
The result that T(1) is open in Q was first proved by Ahlfors [4].
4.6. Closure of the Universal Teichmller Space

For a long time it was a famous open problem, raised by Bers, whether the
closure of T( 1), which is contained in U, actually agrees with U. In 1978,
Gehring [3] showed that the answer to this question is in the negative. He
constructed a counterexample with the help of the simply connected domain
G which is the complement of the curve
y
= {z =
t < co) u {0},
where a > 0 is small (Fig. 6). This 6 is not a Jordan domain, but more than
that, at the origin its boundary is so rigid that G possesses the, following
117
Figure 6.
property: There
of G and uS1 II
not in the closure of T(1).
a positive constant e such that if f is a conformal mapping

e, then f(G) is not a Jordan domain.
For the proof we refer to Gehring [3]. With the aid of this result the
negative answer to the question of Bers is readily established.
Theorem 4.3. The closure of T(1) isa proper subset of U.
PROOF. Let G be the domain delined above and c > 0 the associated constant.
If h is a conformal mapping of the upper half-plane onto G, we prove that S1,
does not lie in the closure of T(l).
Consider an arbitrary point S,,, of the neighborhood
h' S1,II,,
For
w o h' we then have H hG = II S1, hi H
Therefore, eitherf is
not univalent or f is univalent but f(G) = w(H) is not a Jordan domain. It

follows that
is not in T(1).
Recently, Theorem 4.3 has been strengthened: There, exists a confonnal

mapping h: H G, where G is now a Jordan domain, such that
T(1)
(Flinn [1]).
Theorem 4.3 gives rise to the study of the boundary of T(1). If T(1) is
visualized as the collection of quasidiscs
then Flinn's result means that
there are Jordan domains which do not belong to the boundary of T(1).
More information about the boundary of T(1) is provided by some recent
results in the joint paper of Astala and Gehring [1].
We showed in 4.4 that the diameter of U is 12. Since the closure of T(1)
does not coincide with U, we cannot conclude immediately that the diameter
of T(l) is also 12. But this can be proved if we modify slightly the example
which we used for U. For every positive r < 1, the functions w
(w) = w +
nw, w f2(w) = w nw are not only univalent in E = {wI wi> 1} but have
and w w nip, respectively.
the quasiconformal extensions w ' w +
Moreover, an easy calculation shows that
iim(1w12 1)21S11(w)
S12(w)f
If h: H . E is a conformal mapping,
ance formula (1.9) in 11.1.3,
ii S,,01, S120,,
12r.
T(l), and by the invarifi I 2r. It follows that the set
T(1) has diameter 12.
is pathwise conTheorem 2.1 says that the universal Teichmller

nected. Therefore, T(1) and its closure are connected, and by Theorem 3.2,
the set T(1) is even contractible.
III. Universa' Teichmullor Space
118
Since U is not the closure of T(l), the connectedness of T(1) does not imply
that U is connected. In fact, the author just learned of a striking result of

Thurston El] which asserts that U possesses one-point components. (Thurston's result contains Theorem 4.3 as a corollary.)

5.1. Definition ofthe Inner Radius of Univalence
Let A be a simply connected domain of the extended plane whose boundary
consists of more than one point. In 11.2.6 we defined the outer radius of
univalence
a0(A) = sup{I1SJ hAlf univalent in A}
of the domain A. We proved that o0(A) is directly connected with the distance 5(A) of A from a disc (defined in 11.2.1): o0(A) = 5(A) + 6 for all domains A.
Let us now define the inner radius of univalence
univalent in A)
IfS1 hA a
=
of A. Note that a = 0 is always an admissible number, because iis,ir = 0
implies that f is a Mbius transformation and hence univalent. Like the
distance and the radius a0, the inner radius o1 is also invariant under
a1(A)
Mbius
i.e., two Mbius equivalent domains have the same
a1.
The set {S,hf univalent in A} is closed in the family of functions holomorphic in A, when the topology is defined by the hyperbolic sup norm. We
proved this in 4.4 in the case where A was the upper half-plane, and the same
proof applies to an arbitrary A. It follows that we can replace sup bymax in
the definition of o,(A). In other words, if 1IS1JI4 = a,(4), then f is
Theorems 11.4.1 and 11.4.2 imply that o1(A) >0 if and only if A
4uasidisc. As we remarked before, this is an interesting result because quasicon-
formal mappings do not appear in the definition of the inner radius of

univalence.
For a disc,
or,(A) = 2.
This follows directly from Theorem 11.5.2. We shall see in 5.7 that for all
other domains A, the inner radius of univalence is smatter. Before that, we
shall show how to get information about o, from the results derivd in the
previous section. For this purpose, these results must be slightly generalized
so that the special position of the half-plane in the definition of the universal
Teichmller space is removed.
5. Inner Radius of tjnivalence
5.2.
119
Isomorphic Teichmller Spaces
Throughout this section the roles of the upper and lower half-planes in
connection with the universal Teichmller space are as in section 4, i.e.,
is
is conformal in the upper
a self-mapping of the lower half-plane and

half-plane.
Let A be a quasidisc whose boundary contains the points 0, 1, and for

which the direction from 0 to I to orients
positively with respect to A.
The universal Teichmller space TA is defined by means of quasiconformal
self-mappings of the complement of A which fix 0, 1, and
exactly as we did
it in 1.1 in the'case of a half-plane. Theorem 1.2 can be proved for A word for
word as in the case of a half-plane. It follows that the points of TA can be

represented by conformal mappings of A which are restrictions to A of
quasiconformal mappings of the plane and which fix the points 0, 1, and
The given quasidisc A can be regarded as a point of the universal Teichmuller space T = TH. By this we mean that there is a unique point p e TM such
that
= A whenever
p (cf. 1.5).
Take a fixed mapping
with the property
= A, and consider the
transformation
(5.1)
This is an isomorphism between

and TA in the sense that it is a bijective
isometry. Obviously, (5.1) is well defined in
and a bijection of TM onto TA.
o
if
J
= f2 oft', and so (5.1) preserves Teichmuller distances.
Note that under (5.1) the origin is shifted: the point [Mo] TH maps to the
origin of TA. More generally, [p] maps to the point represented by the
complex dilatation of
Suppose
TA. In order to study the mapping [p] ' S114. we define
as the' space of functions p holomorphic in A for which the norm

is finite. As in the case A = Ii, we set U4 = {p =
univalent in A); by Theorem 11.2.3, this is a subset of Q4. Finally, T4(1) = {S, UA)f
has a quasiconformal extension to the plane). Our previous space Q will now
be denoted by Q11.
The function w
is meromorphic in A if and only if f is meromorphic in H. From HS.JA = IS1 SJ IIH we get
IIS,h
and conclude that
the mapping
+ IIS,1.011K
is finite if and only if uS,
is
finite. It follows that

(5.2)
S1 .
is a bijection of Q1 Onto Q4. Clearly, (5.2) maps UM onto 114 and T11(l) onto
T4(1). If w1
i = 1,2, then by formula (1.9) in 11.1.3,
SW1
iiA
H,1.
III. Universal Teichznller Space
120
onto QA is also an isometry.
We see that the mapping (5.2) of
Let
[ii) .
the mapping
be the mapping [p] -+ S111, of TI, onto T,,(l) and
the same as
of T4 onto TA(l). Then (5.1) followed by
followed by (5.2)
5.3. Inner
Radius and Quasiconformal Extensions
Let A be a quasidisc. We define the spaces QA. UA, and TA(l) as in 5.2.
Let h: H p A be a conformal mapping. Generalizing (5.2) we can define a
bijective isometry of Q,, onto QA by
Sf
We
(5.3)
4 S101,
conclude that for varying quasidiscs A, all spaces
and TA(l) are
isomorphic. In particular, Theorem 4.2 can be generalized:

For all quasidiscs A,
TA(1)=intUA.
(5.4)
This relation leads to a new characterization of the inner radius of univalence. It follows from the definition of c, and from the subsequent remark
about sup and max in the definition that the closed ball
E QA II
HA
c,(A)} is contained in UA. By (5.4), the interior of this ball lies in TA(l). In
other words, if
ItS1 11.4 <c,(,4),
then f is not only infective in A but has a quasiconformal extension to the plane.
This result generalizes the statements of Theorems 11.5.1 and 11.5.2 which
are concerned with a disc. It sheds new light on the inner radius of univalence
and explains why quasiconformal mappings play a role. We conclude that

the definition of
inner radius of univa!ence can be expressed in the form
a,(A) = inh{
S1 II A
univalent in A, f(A) not a quasidisc}.
(5.5)
This characterization yields upper bounds for c,(A).

Another way of expressing (5.5) is that B(0, c1(A)) =
(Ic' (IA <o,(A)}

is the largest open ball in QA centered at the origin which is contained in
TA(l). The inverse of the isomorphism (5.3) takes this ball onto the ball
B(Sh, c1(A)) of Q,,. This gives a characterization for the inner radius of uni-
valence in geometric terms: If h is a conformal mapping of H onto a quasidisc

A, then u,(A) is the distance from the point 5h to the boundary of TH(l).
This result gives additional information about the mapping [ii] S111, of
T,, onto T,,(1). We proved in section 4 that
is open in
Now we can
express this result in a more precise form:
Theorem 5.1. The largest open ball of
and lies in TI,(l) has the radius
which is centered at the potut S11,
121
We recall that the domain constant t5also has a similar geometric interpretation: (5(A) is the distance from the point S,, to the origin of TH(l). The
geometric interpretations of and (5 can actually be exploited. Before doing
it we shall estimate the inner radius for
specific domains. We first

derive a lower estimate for a,(A) with the aid of quasiconformal reflections.
5.4. Inner Radius and Quasiconformal Reflections

The outer radius of univalence can often be readily determined thanks to the
relation
= (5 + 6. In contrast, there seems to be no easy way to find the
exact value of the inner radius of univalence of a quasidisc. A modification of
the proof of Theorem 11.4.1 yields a lower bound for cr1(A) which turns out to
be sharp in certain cases.
Lemma 5.1. Let A be a bounded quasidisc which can be exhausted by the

domains A, =
0 < r < 1, and). a quasiconformal reflection in (5A
which is continuously
off 3A. Let A'

a,(A)
inf
zi
Then
12(z)I
(5.6)
Let f be a meromorphic function in A. Suppose first that f is holomorphic on

As in the proof of Theorem 11.4.1, we setf w1/w2 and write
PROOF.
(A(z)
(z)
+ (A(z) z)w(z)
Let g = o
and =
be the complex dilatation of g. The reasoning
in the proof of Theorem 11.4.1 shows that if
< 1, then f is univalent
and g is a quasiconformal extension of f (In proving Theorem 11.4.1 we

resorted to inequality (4.4) to determine the boundary values of g and (5g.
Here we can proceed more directly, since we may assume that the right-hand
expression in (5.6) is positive.)
Now
=
1
Since
is
a sense-reversing quasiconformal mapping,
away from I in absolute value. It follows that

< I.
<
Direct computatiOn gives
&p(z)
We conclude that if
if
< 1 if
is bounded
=
if
z)2Sf(z)/2
if and only if
122
111. Universal TeichniUfler Space
+ (2(z) z)2S1(z)/21
t < 1. A fortiori, this is the case if
IS1(z)I 2
I2(z)zI 2
This in turn holds whenever

I 1(z)I
IISJIIA<2 inf
zL ?74(Z)
(5. )
Under this condition S1 is a point of the ball B(O, o,(A)) of QA.
we apply the above reasoning to fI A, for

1ff is not holomorphic on
suitable values of r tending to 1. The function z rA(z/r) is a quasiconformal
It follows that whenever (5.7) holds,
reflection in 34,., and
=
the mappings 114, are univalent and have quasiconformal extensions with
arguuniformly bounded maximal dilatations. As in 11.4.3, a normal
ment then shows that under (5.7), S16 TA(l), and (5.6) follows.
0
Remark. Let h be a Mbius transformation, q
holoh', and
= h(z).
Then
132(z)l
12(z)
k'(C)
This can be verified by direct computation (cf. 11.4.1). It follows that Lemma
5.1 holds for quasidiscs which are Mbius equivalent to a quasidisc A fulfilling the conditions of the lemma.
Let us test the accuracy of (5.6). If A is the upper half-plane and 2(z) =
then
131(z)I 131(z)I
12(z) z12,A(z)l
Hence (5.6) gives o,(A) 2. We get the same result if A is the unit disc and
1(z) = 1/i. Consequently, in these two cases the lower bound in (5.6) is equal
to a,(A).
If A is the exterior of the ellipse {z = e' +
k < 1,
+ k/w with 2w z + (z2
we have the reflection z '
Now (5.6)
gives the simple estimate o,(A) 2(1 k)2. This lower bound is asyrnpktically correct as k +0 or k ' 1, but is not sharp, as we shall see in 5.6.
5.5. Inner Radius of Sectors

Lemma 5.1, combined with the characterization (5.5), makes it possible to
determine r,(A) for sectors (Lehto [7], Lehtinen [2]).
123
5., Inner Radws of Univalcuce
Theorem 5.2. Let A be the sectoral region {zlO < argz <kir}, 0 < k < 2. Then
ifO<k1,
"12k2
(5.8)
<k<2.
PROOF. A continuously differentiable quasiconformal reflection
be
in
can
defined by the formula

A(z)
zeA.
It follows that
2(z)I IA(z)I 1/k
Setting
re'9
....
II 1/ki
z12
z12
we obtain
z
sin(O/k).
In addition,
2klmzim
i(z) =
IzI Ilki
= 2kr sin(O/k).
Therefore,
Thus (5.6) gives, in view of the Remark in 5.4,
o,(A)2k(l
(5.9)
II),
which is (5.8) with > instead of equality.
We stilt have to show that equality holds in (5.9). Suppose first that
0 < k 1. In this case it is easy to prove directly, without making use of
Lemma 5.1, that o1(A) = 2k2. The proof is based on the fact that
is
equal to the distance from
to
where h: H * A is a conformal
mapping. Now z , h(z) = 2k is such a mapping. Hence, Sh(z) = (1 k2)/(2z2)
and AShIIU = 2(1 k2). For the function z -+ g(z) = Iogz we have S9(z) =
1/(2z2), and we know that S9 is not in TH(l),
because g(H) is
not
a Jordan
domain. It follows that

IS9 SkIIH
2k2.
In conjunction with (5.9), this yields the first result (5.8).

As we said, in this case the inequality (5.9) is readily obtained directly. In
fact, if p is a boundary point of TH(l) nearest to Sk, then
G1(A) = 114' ShII 1Q11
SinceQisnotin TH(1),wehave lkoll 2,andso
IISkII.
124
UI. Universal TeichmUller Space
1,
Figure 7. SIeTA(1) closest to the origin.
o,(A) 2
2(1
k2)
2k2.
no nearest point exists, an

c-reasoning yields this inequality.)
In order to complete the proof, we still have to show that a,(A) 4k 2k2
if I <k <2. In view of (5.5), this follows if we find a conformal mapping
f: A * B such that
= 4k 2k2 and B is not a Jordan domain.
(If
Set
{zljarg(l z)I <kit/2}.
B = {zf Iargzl <kn/2}
This is not a Jordan domain because of the behavior of 3B at 00 (Fig. 7). The
Schwarzian derivative of the conformal mapping w1: H ' H B, with w1(0) =
1, w1(cc) = 0, w1(1) =
can be computed (see Nehari.[2], p. 203). it follows
that
4k2
Sz2
2kk2
2(z
if
k22k
2z(z
If
A conformal mapping w: H -+ B is obtained if the map z w1(z2), defined in

the first quadrant of the plane, is reflected in the imaginary axis. The composition rule for Schwarzian derivatives yields
ik2
=
With h(z) =
uS1
2z2
the function f =
II,,, we
4k2k2
+ (2
h'
maps A conformally onto B. Since

finally arrive at the desired result
uS1 hA = sup
wa
4y2(4k 2k2)
2
22
y 1)2 +4xy
2
4k
2k 2
This completes the proof of (5.8). The example of the conformal mapping
f: A . B is due to Lehtinen [2].
125
5. Inner Radius of Univalenoc
5.6. Inner Radius of Ellipses and Polygons

In many cases Theorem 5.2, in conjunction with the fact that the hyperbolic
metric depends monotonically on the domain, provides an efficient means for
deriving estimates of the inner radius. The underlying ideas, which are due to
Lehtinen ([5]; cf. also [41), are in the following two lemmas.
Lemma 5.2. Let A be a quasidisc which is contained in a domain B4 Mbius
equivalent to the sector A4 =
<kirf2}. If 0 < k 1, assume that a
vertex v of B4 lies on
Then
ey,(A) 2k2.
If 1 <k < 2, assume that there are points
and z2 in 3A such that for a
Mbius transformation g mappuzg B4 onto A4, g(z1)
g(z2) =
Then
a,(A) 4k
2k2.
PROOF. Suppose first that 0 < k 1. Let g be a Mbius transformation

mapping B4 onto A4 with g(v) = 0. Set f(z) = logg(z). Then f(A) is not a
quasidisc, and so by (5.5), o,(A) IS, 114. By the monotonicity of the hyperbolic metric (formula (1.2) in 1.1.1), IS, hA II S1
= 2k2.
Suppose next that 1 <k < 2. From the proof of Theorem 5.2 we deduce
k2 and
that
= f(e"2) = ao. Then f(g(A)) is not a Jordan domain, and by
reasoning as in the case 0 <k 1, we arrive at the desired estimate.
0
the existence of a conformal mappingf of A4 such that IS, IlAk = 4k
The second lemma, which does not rest on Theorem 5.2, is more general.
Lemma 5.3. Let A be a quasidisc. if every two-point subset of A is contained in
the closure of a quasidisc B Afor which o1(B)> m, then
c71(A) m.
PROOF. Let an e > 0 be given. There exists a meromorphic function fin A for
which IIS,114 <
+ c but which is not univalent. Let
and 22 be two
different points of A such that f(z1) = f(z2), and B A a quasidisc such that
{z1,z2}
B and o,(B)> m. Since eitherf is not univalent in B or else f(B)is
not a quasidisc, II S1 > o1(B). By the monotonicity of the hyperbolic metric,
Hence 1(A) > m e, and the lemma follows.
II 5, 114
As a first application, let us reconsider the case in which A is the exterior

of the ellipse {z = e" + ke"jO p < 2ir},0 k < 1. The domain A is contained in the infinite domain B whose boundary consists of two circular arcs
of which one passes through the points 1 + k, 1(1 k), (1 + k) and the
other through 1 + k, i(1 k), (1 + k). Since B is Mbius equivalent to
126
the angle A, with I =
+ (4/n)arctan k, Lemma 5.2 yields the upper estimate
2 (32/ir2)(arctan k)2.
On the other hand, simple geometry shows that each pair of points in A
lies in a domain B' c A Mbius equivalent to the angle A,. with I' = I + (2/n)
arctan(4k/((1 k)(k2 + 6k + 1)1/2)). By Lmnia 5.3,
aj(A) 2 (8/ic2)(arctan(4k/((1
k)(k2
+ 6k +
This is sharper than the simple estimate o,(A) 2(1
k)2
1)h/2)))2.
we obtained from
Lemma 5.1.
In the second application of Lemmas 5.2 and 5.3, we suppose that A

is a finite polygonal domain with interior angles k,n, I = 1, 2, ..., n. Then
Lemma 5.2 yields immediately the upper estimate
(5.10)
1k, II)}.
In certain cases, we can refine this result with the aid of Lemma 5.3 so as
to obtain the exact value of the inner radius. If A is a triangle or a regular
n-sided polygon, then (5.10) holds as an equality.
It follows that fora triangle A,
where
(5.11)
is the smallest angleof A, and for a regular n-sided polygon

2)2
og(A) =
2("
(5.12)
For the proofs we refer to Lehtinen [5]; the results (5.11) ahd (5.12) are also
due to Calvis [1].
5.7. General
Estimates for the Inner Radius
Let h: H -. A be a conformal mapping. Then () is the distance from Sh to

the origin of
and u1(A) the distance from Sh to UH\TH(l). The set TH(l)
contains the ball B(O, 2), and U,, is contained in the closure of B(0, 6). It
follows that the double inequality
2 () + o,(A) 6
holds for all domains A conformally equivalent to a disc.
The left-hand inequality o,(A) 2 5(A) is an equality for all sectors

A = {zfO < argz <kn}, 0 < k 1, because by Theorem 5.2,
= 2k2,
and we have earlier computed 5(A) = 2(1 k2). Other extremals are obtained as follows. Consider a domain for which II fl = 2 and which is not a
quasidisc. If 0 < r < 1 and S, = rSh, then for A = 1(H), we have 5(A) = 2r
and ci1(A) = 2(1
r).
127
5(A) is of interest only if () is close to 6.

The upper estimate r,(A) 6
For
< 4, a better estimate can be derived.
Theorem 5.3. For all domains A con formally equivalent to a disc,

. 2.
Equality holds
t5.13)
and only if A isa disc.
PROOF. Let A be
an arbitrary quasidisc. Every Jordan domain is Mbius
equivalent to a subdomain of H having 0 and as boundary points. We may

assume, therefore, that A itself is such a domain.
In A, we consider the function z
logz, for which S1(z) l/(2z2).
From the monotonicity of the hyperbolic metric it follows that S,. 'A 2.
Because f maps both 0 and
to infinity, f(A) is not a Jordan domain.
Hence, we obtain (5.13) from the characterization (5.5) of the inner radius.
The same idea, in a refined form. can be used to prove that
2 only
if A is a disc (Lehtinen [2J). We now assume that A c H has two unite
boundary points on the real axis. If A is not H,
are two finite points, in
such that he open interval on between these points lies in the
complement of
A simple geqmetric argument shows that A then lies in a
non-convex sector both of whose sides contain a point of t3A at an equal
distance from the vertex (for the details, see Lehtinen [2]). Therefore, we may
assume that A lies in an angle Ak = {z0 <arg: <kir}, 1 <k < 2, such that
the points I and
are on the boundary of A.
Instead of the logarithm, we now consider the extremal mapping f of the
sectoral domain Ak exhibited in the proof of Theorem 5.2. Since f(I)
the image of A under fL4 is not a Jordan
It follows from (5.5), the
monotonicity of the hyperbolic metric, and Theorem 5.2, that
c,(A) IISfIAIIA <

By Theorem 5.3, every j-oint
boundary of
= 4k 2k2 <2.
0 of
has a distance < 2 from the
ft is an open question what values the inner radius of univalence can

assume for K-quasidiscs. The secloral domains show that
inf{a1(A)IA K-quasidisc}
We may also ask whether Theorem 11.4.1 holds if e(K) is replaced by
CHAPTER IV
Riemann Surfaces
Introduction to Chapter IV
A number of textbooks have been written on the subject of Riemann surfaces.
In spite of this, we found it advisable to include in our presentation a chapter
in which we have collected the material on Riemann surfaces that will come
into play in Chapter V. A brief survey of the general theory of Riemann

surfaces is given in sections 13 and of groups of Mbius transformations in
section 4. We have occasionally lingered on some topics slightly longer than
would be strictly necessary for later needs, in order to provide the reader with
a broader background.
In section 1 standard definitions of manifolds and Riemann surfaces and
of functions and differentials are given. We have also treated in some detail
the classical problem of Gauss to map a portion of a surface imbedded in

euclidean three-space conformally into the plane. This problem inaugurated
the theory of quasiconformal mappings around 1825. It also gives a first hint
of the intrinsic role of quasiconformal mappings in the theory of Riemann
surfaces.
Section 2 deals with covering surfaces and their topology. The main results
are formulated but proofs are usually only sketched. References for complete
proofs are to the monograph Abhors and Sario [1].
In section 3, results of section 2 are applied to Riemann surfaces. In conjunction with the general uniformization theorem, they yield the fundamental
result that, modulo conformal equivalence, every Riemann surface is the

quotient of a disc or the finite plane or the extended plane by a discontinuous
group of Mbius transformations. The section concludes with a study of how
mappings between Riemann surfaces induce mappings between the covering
surfaces and the covering groups.
129
Section 4 provides a survey of some main features of the theory of groups

of Mbius transformations. For detailed proofs, reference is usually made to
the monograph of Lehner [1]. In section 5 we have collected various results
on compact Riemann surfaces, using Springer [1] as a reference.
Quadratic differentials play a remarkable role in the theory of Teichintiller
spaces. Therefore, the geometry and the metric induced by a quadratic differential are studied quite extensively in sections 6 and 7. Here we have largely
utilized the recent monograph of Strebef [6), to which we also refer for more
details.

1.1. Real Manifolds
A real n-dimensional manifold M is a Hausdorif space with a countable base
for topology which is locally homeomorphic to the euclidean space W'. This
means that to every point p M there is an open neighborhood U M of p

and a homeomorphism Ii of U onto an open set in R.. Such a mapping h is
on M. A set of local parameters is said to be an atlas
calhd a local
of M if the unlos oftheIr domains covers M.
Let h1, 112 be local parameters on M such that their domains U1 and U2
have
Intersection. Restricting h1 and h2 to U1 n U2, we obtain a
homecnl*phlc mapping h2 o
between the open sets h1(U1 n
and
U3) In R1. With the help Of this induced mapping, which we call a
lrwuformation, properties definable in can be transported to the
manifold M.
An alMa Is called differentiable if for all pairs of local parameters, the
parameter transformations are differentiable where defined. A maximal differentiable atlas is called a differentiable structure on the manifold, and a
manifold with a differentiable structure is a dWerentiable manifold. When
speaking in the following of the local parameters on a differentiable manifold,
we always assume that the parameters belong to its differentiable structure.
With obvious modifications, the method used for defining differentiable
manifolds leads to manifolds with other structures. For instance, if every
parameter transformation is of class Ck (has continuous partial derivatives
up to order k), we speak of
Or if the word "differentiable" is
replaced by "real-analytic", we obtain real-analytic manifolds.
A continuous map f of a differentiable manifold M into a differentiable
manifold N is said to be differentiable at a point pe M if there are local

parameters I, and k, defined in neighborhoods of p and f(p), respectively,
such that the composition k of a h1 is differentiable at h(p). 1ff is differentiable at p. then k of o
is differentiable at h(p) for all local parameters h and
IV. Riemann Surfaces
130
k near p and f(p). The map f: M
is said to be differentiable if it is
differentiable at every point of M.

A path on a manifold M is a continuous mapping of an interval into M. If
M is differentiable, we can speak of a differentiable path.
Let f be a mapping of an n-dimensional Ct-manifold M into a manifold N.
The function f is said to be (Lebesgue) measurable on M if for every local
parameter h on M defined in an open set U, the function .10 n is measurable
with respect to n-dimensional Lebesgue measure, i.e., if for every open set
G
N. the inverse irvage (f" 'r'r'(G)=
is a measurable
Borel set,f is Borel
is always
subset of h(U). If the preimage Ii(U
measurable on M.
latter definition can be used for au manifolds, because
Borel sets are preserved under homeomorphisms.
A surface is a connected two-dimensional manifold. A surface is aiways

pathwise connected. This follows by standard reasoning from the facts that a
A surface may
surface is connected and, clearly, locally pathwise
literature a compact surface is called
or may not he compact. In the
closed, a non-compact 'urface open.
1.2. Complex Analytic Manifolds

If we replace in the definition of an n-dimensional manifold in 1.1 the euclidean
space
by the space C" of n-tuples of complex numbers, we get a complex
n-dimensional manifold. The euclidean space p2n becomes identified with
if we associate
with (x1 + iy1,...,
the space
+ iy1')e C". Therefore, a complex n-dimensional manifold can be regarded
as a real 2n-dimensional manifold.
Let M be a complex n-manifold. Suppose M has an atlas in which all
parameter transformations are biholomorphic, i.e., aleng with their inverses
they are holomorphic functions of n complex variables. Maximal atlases with
this property are called complex analytic structures on M. We say that a
manifold equipped with such a structure is a complex analytic
Definition. A one-dimensional connected complex analytic manifold is a
Riemann surface.
In the case n 1 we also call the complex analytic structure conformal.

Thus a Riemann surface is a surface with a conformal structure.
Besides pl3ying a central role in complex analysis, the notion of a Riemann
surface has idiiiated or influenced a multitude of other mathematical disciplines.
observation that the natural habitat of an analytic function
of the complex plane but a surface which is locally

equivalent to a plane domain appears already in Rieinann's
thesis
1851. A rigorous definition of Riernann surface in modern
terms was given as early as 1913 by Weyl [1]. Weyl's monograph also
is
contains the first precise definition for a surface.
131
A domain in the complex plane is to be regarded as the Ricmann surface

with its natural conformal structure induced by the identity mapping, unless
otherwise stated. The Riemann sphere is another example of a Riemann

surface with a natural conformal structure.
An open subset ofa Riemnn surface S is assumed to have the conformal
structure induced by the conformal structure of S. When speaking of the local
parameters on a Riemann surface, we always assume that the parameters
belong to its conformal structure.
Analytic functions can be defined on Riemann surfaces. A continuous map
f of a Riemann surface S into a Riemann surface W is analytic at a point p e S
if there are local parameters h and k, defined in a neighborhood of p and f(p),
such that kofolr' is holomorphic at h(pl. This is an invariant definition,

independent of the choice of the local parameters near p and 1(r). if f is
analytic at each point, I is an analytic mapping. In case W is the complex
plane, an analytic mapping f is said to be holomorphic, while if W is the
extended plane, f is termed meromorphic.
An injective analytic function on S is called a conformal mapping of S.
It follows fro.m the definitions that local parameters of S are conformal

mappings of open subsets of S into the complex plane.
1.3. Border of a Surface

A bordered surface is a connected Hausdorif space with a countable base for
topology in which there exists an open covering by sets homeomorphic

with sets open in the closed half-plane H
{x +
O}. The concepts of a
local parameter and an atlas can be introduced in an obvious manner for

bordered surfaces.
Let
be a bordered surface, p E S*, and h a homeomorphism of an open
neighborhood of p onto an open subset of H. Let us assume that h(p) is an
interior point of H. If k is another homeomorphism of a neighborhood of p
onto an open set in H, we can choose a disc D with center at h(p) and with
closure in H, such that kol(' is homeomorphic in D. Now koh' is an
injection of D into the plane and it is continuous in the topology of the plane.
Since D is an open neighborhood of h(p) in the topology of the plane, it

follows from the invariance of open sets under continuous injections that
k(p) must be an interior point of H. Hence, if the image of p under one local
parameter of S is an interior point of H, then it is so under all local parameters of S* defined 'at p.
It follows that we can write
S is the subset of S* whose
S
points map into the interior of ii under any local parameter, and B the set
whose points map on the boundary of H. Clearly, S is open in S, and so B
contains points of S.
is closed. Each neighborhood of every point of
Consequently, S* is the closure of S. It is easy to show that S is connected,
whence it follows that S is a surface. The set B is called the border of S*. Each
pointpEli has a neighborhood U in
which is homeornorphic to the
132
half-disc {z = x + iyIIzI < 1, y O} under a parameter which maps p to the

origin and B U onto the line segment ( 1, 1). We conclude that B is a real
one-dimensional manifold.
A bordered Riemann surface is a bordered surface S* = S u B with a con-
formal structure on S determined by an atlas' on
S is a
Riemann surface.
1.4. Differentials on Riemann Surfaces

Let S be a Riemann surface whose conformal structure is determined by the
local parameters; with domains
i = 1, 2
a
complex-valued
holomorphic
function on S. Suppose that z, and
be
Let f
= fo
have overlapping domains, and write f, = fo
It is customary to regard z, and Zj as complex variables. If we do so and differentiate
=
we arrive at the invariance
(1.1)
=
It follows that while an invariant derivative cannot be defined for f, we can
speak of the invariant differential df, defined locally by (1.1).
Let us now generalize the notion of a differential. A collection p of

defined on U1, i = 1, 2, ..., is said to be an
complex-valued functions
(m, n)-differential on S if
dz. m
St
(1.2)
in U1 n
The function element is a representation of in terms' of the
local coordinate z1. The differential qs is said to be holomorphic if all functions
are holomorphic. Meromorphic differentials are defined similarly.

and on S are equal if their local representations
in the same local coordinate always agree. In case and
are meromorphic, we conclude t(iat = if their representations agree for some local
coordinate.
From the definition it is clear that the set of (m, n)-differentials on S and
the subset of all holomorphic (m, n)-differentials form linear spaces over the
complex numbers. We can also form the product of an (m, n)-differential and
a (p, q)-differential in an obvious manner and so obtain an (m + p. n + q)differential.
Particularly important in the following is the case m = 2, ii = 0. Then q, is
called a quadratic
Such differentials will be studied in sections 5,
6, and 7 of this chapter, as a preparation for the applications in Chapter V. A
holomorphic or meromorphic (1,0)-differential is called an Abelian differTwo (m, n)-differentials
ential. Its square is a quadratic differential. Deeper connections between

holomorphic Abelian and quadratic differentials will be investigated in 5.5.
i. Manifolds and Their Structures
133
Quasiconformal mappings between Riemann surfaces will be introduced

in V.1. It turns out that in this setting the complex dilatation generalizes to a
( 1, 1)-differential. For a ( 1, 1)-differential q it follows from the invariance
(1.2) that we can speak of the function
on S. if is Lebesgue measurable
<1, then the ( I, 1)-differential q is called
in all local coordinates and II
a Beltrami
In applications (1, 1)-differentials are often important, because they can
be integrated with respect to the two-dimensional Lebesgue measure. This
follows from the fact that if z w is a change of local coordinates, then
=
where Iw'i2 is the Jacobian of the mapping z * w. If p is a quadratic differ-
Thus the
for all local representations
ential, then o1 dzl I = I I dz,
absolute value of a quadratic differential is a (1, 1)-differential. Another important observation is that the product of a quadratic differential and a
(
I, 1)-differential is a (1, 1)-differential.
1.5. Isothermal Coordinates

The natural question of how to make a concrete surface in
into a Riemann surface leads us to quasiconformal mappings. Let S be an orientable
C'-surface in R3, and f =
the inverse of a local parameter of S. The
metric on S is defined locally by the line element ds, where
ds2
tiy
= Edx2 + 2Fdxdy + Gdy2.
(1.3)
Here
i-1
the classical Gaussian quantities. The expression (1.3) is invariant, i.e.,

independent of the choice of the local parameter.
Using the complex notation dz = dx + i dy,
= dx i dy, we obtain
are
from (1.3)
ds = Al dz +
with
E+
+ 2 JIG F2
(1.4)
IV: Riemann Surfaces
134
It is a classical result (and not difficult to prove by direct calculation) that

conformal in the sense that it preserves angles if and only if E G, F = 0.
This condition is equivalent to being identically zero. In this case
is
ds=)jdzj.
Local coordinates z of S with this property are called isothermal.
Let us consider another local parameter of S which defines the local coordinates w. If the coordinates z and w are both isothermal (ds =
and
if the induced mapping z ' w is defined in some non-empty open set of the
plane, then
=
This shows that z , w is conformal or indirectly conformal. Since S is on-
eatable, we may assume

2 + w is conformal. We conclude that isothermal
coordinates define a natural conformal structure for an orientable C'-surface,
which thus becomes a Riemann surface.
1.6.
Riemann Surfaces and Quasiconformal Mappings
We are thus led to the

problem of finding isothermal coordinates for
a surface. Without bothering about minimal conditions, we show how a
solution is obtained with the aid of the existence theorem for Beltrami
equations.
Theorem 1.1. Ecery orientahk' C2-surfiwe in
can be made into a Riemann
surjace.
Let S be an onientable C2-surface. Consider an arbitrary local parameter of S inducing local coordinates z in a domain A of the complex plane.
The theorem follows if we can transform the z-coordinates diffeomorphically
so that the new coordinates are isothermal.
Expressed in terms of z. the line element of S is of the form (1.4). Here jz
is continuously differentiable and by (1.5), we have
< I in every
relatively compact subdomain of A. Let z - w be a quasiconformal mapping
of such a subdomain with complex dilatation
By the Existence theorem
1.4.4 such a mapping w exists, and by the remark in 1.4.5, w is continuously
differentiable and
0 everywhere. Comparison of
PRooF.
Idwi =
with (1.4) shows that

A
ds = ---Idwl,
We
see that the w-coordinates are isothermal, and the theorem is proved.
of Covering Surfaces
135
In essence, Theorem 1.1 is due to Gauss [1]. Since every sense-preserving
diffeomorphism is locally quasiconformal, there is no undisputed criterion

to determine the first appearance of quasiconformal mappings in analysis.
realized the importance of
But in developing the theory
finding locally injective solutions for a Beltrami equation (i.e., quasiconfqrinal
mappings with a given complex dilatation) and he actually constructed such
solutions. Therefore, it is not without justification. to say that quasiconformal
mappings entered analysis around .1825, in connection with the problem of

how to map a plane domain conformally onto a portion ofa surface
in euclidean three-space.
For Gauss conformal mappings were just a tool in differential geometry. It
was Riemann who recognized the fundamental connection between conforina!
mappings and complex analysis. Later, the concrete method of Theorem 1.1
to generate Riemann surfaces was used by Klein [1], whose work paved the
way fOr Weyl's monograph [1] cited in 1.2.
Theorem 1.1 and its proof mark the first indication of the intimate relationship between' Riemann surfaces and quasiconformal mappings. Later we
shall uncover plenty of additional evidence of the depth of this
and Theorem 1.1 will be generalized in various ways. -
2.
of Covering Surfaces
2.1. Lifting of Paths

The unifying link between the theory of abstract
surfaces and complex analysis in the plane is provided by covering surfaces. In this section, we
shall discuss.topological properties of covering surfaces.
A smooth covering surface of a surface S is. a pair (W,f), where W is a
surface and f: W , S is a local homeomorphism. The mapping I is called a
projection, and, the inverse images of a.point peS are said to lie over p. Being
locally a homeomorphism, f is both continuous and open.
Let y be a path on S, more precisely a continuous map of the closed unit
interval I =
l} into S. A path yI on W with initial point a = y'(O)
and with' the property foy' = y.is called a
of y fvom a. It is easy to prove
that on' a smooth covering surface the lift from a fixed, 'inital point, is unique
(AhlforsSario [1], p. 28)...
From the local injectiveness off it is clear that a part of y can always
lifted from
arbitrary point a lying over its initial point. the wbole of y
cannot be lifted, there is a 0 < t0 1, such that y restricted to any closed.
subinterval of [0, t0) can be lifted from a but vi [0, to.] cannot be so lifted.
A smooth covering surface (W,f) of S is said to be unlimited (regular, in the
terminology of AhlforsSario [I]) if every path on S has a lift to Wfrom each
136
point lying over its initial point. In this case f: W . S is surjective, and the
is the same for all points peS.
cardinality of the set!
Unlimited covering surfaces have an important topological property

(AhiforsSarlo [I], p. 30).
Theorem 2.1 (Monodromy Theorem). Let (W,f) be an unlimited covering

Yo and Yi homotopic paths on S. Then the lifts of
and on Wfrom the same initial point have the same terminal point and they
are homotopic.
surface of a surface S. and
Suppose, in particular, that the surface S is simply connected, i.e., that the
fundamental gioup of S is trivial. In this case the monoclromy theorem yields
an interesting corollary:
If (W,f) is an unlimited covering surface of a simply connected surface S,
then the mappingf: W p S is a homeomorphism.
For since the projection f is continuous, open and surjective, it is enough

to show that f is injective. Assume that there are two points a and of W.
such that f(a) = f(b). A path y from a to b then has a projection on S which
is a closed curve. This is homotopic to zero, since S is simply connected. By
the monodromy theorem, y terminates at the same point as the constant path
t
a. Hence a = b.
2.2. Covering Surfaces and the Fundamental Group

Let S be a surface, (W,f) an unlimited covering surface of S, and a a point of
S. We
F. thfundamental group of S whose elements are homotopy classes [y] of closed paths y on S from a. Fix a point a' W over a, and
consider a homotopy class [y] containing a path whose lift from a' is closed.
It follows from the Monodromy theorem that the lifts of all paths of [y] from
a' are then closed, and such elements [y] form a subgroup of FQ. We denote
this subgroup, determined by the triple (W, f, a'), by F'4..
The choice of a e S is immaterial. For given another point b S, consider a
path a on S from a to b; let b' be the terminal point of the lift of a from a'. If
then fy] -+
carries
[c'ya]
is a group isomorphism of F. onto Fb which
onto the subgroup F'b. determined by (W,f, b'). Also, if a' e W is
replaced by another point a" of W over a, then the groups determined by

(W,f,a') and (W,f,a") are conjugate subgroups of Fa. This Can be easily
verified.
An unlimited covering surface (W,f) of S is said to be normal if the triple

(W,f, a') determines a normal subgroup of F4. This notion is well defined,
because the property of a covering surface being normal does not depend on
137
W. If (W,f) is normal, then all triples

and
the choice of the points
(14'/, a') with a' ef' {a) determine the same subgroup of F4.
There is an important connection between the topologies of W and S:
Let (W, f) be an unlimited covering surface of S. Then the fundamental group
of Wis isomorphic to the subgroup of F4 determined by a triple (W, f, a').
The proof is easy. We consider closed paths)' of W from a' and check that
[y] [Joy] is a required isomorphism (AhlforsSario [1], p. 36).

The connection between the triples (W,f, a') and the subgroups ofF4 makes
it possible to partially order the unlimited covering surfaces of S according to
strength. The strongest covering surfaces are those which determine the trivial
subgroup of F4. They are called universal covering surfaces of S. By what
just proved, a universal covering surface is simply connected. It is of course a
normal covering surface.
Every surface possesses universal covering surfaces. This can be shown by
direct construction. Given a surface S and a point a S, consider all paths y
of S from a to a point p. We define the set
W= {p'
and the mapping f: W * S by the requirement f(p') p. Then a topology
can be introduced on Wso that (W,f) is a universal covering surface of S. For
the details of the proof we refer to AhlforsSario [1], p. 35. There the more
general result is proved that, given any subgroup of F,, there exists ar unlimited covering surface of S which determines this subgroup.
The notion of a universal covering surface is due to H. A. Schwarz, who
noticed its importance in the theory of Riemann surfaces (see Theorem 3.4 in
the next section).
2.3. Branched Covering Surfaces

In elementary function theory, Ricmann surfaccs are first encountered in
connection with the mapping z -+ z4. This defines the plane as a covering of
itself, but in such a way that the projection mapping has branch poin.ts at 0
and cc. More generally, if f: W -+ S is a non-constant analytic mapping

between the Riemann surfaces W and S,.then (W,f) need not be a smooth
covering surface of S. This state of affairs leads us to generalize the notion of
smooth covering surface.
A covering surface of a surface S is a pair (W,f), where W is a surface,

f: W + S is a continuous mapping, and every point p E W has a neigh-
borhood U such that (U\(p},ftU\{p}) is a smooth covering surface of

S\ {f(p)}.
A smooth covering surface is trivially a covering surface. A covering surface
which is not smooth is called branched.
IV. Riemann SurFaces
138
We can deduce from the definition that the projection mapping f: W S

behaves locally like the mapping z -. z" for suitable n. More precisely, the
following result is true:
Lemma 2.1. Let (W,f) be a covering surface of S. For every peW, there are
parameter discs U p and f( U), with local parameters k and h normalized by
k(p) = h(f(p)) = 0, such that in U,
(2.1)
where tz is a natural number.
The proof is given in AhlfQrsSario [1), p. 40. Conversely, if f: W-+ S is a

continuous mapping and the above condition holds, we conclude immediately
that (W, f) is a covering surface of S. Thus this condition characterizes

covering surfaces.
We mention here that there are other non-trivially equivalent characterizations of covering surfaces, even thQugh we shall not be using them in what
follows. Let S and W be surfaces and f: W S a continuous mapping. Then
(W,f) is a covering surface of S if and only if f is locally homeomorphic with
the possible exception of a discrete set, or if andonly 1ff is light (the preimage
of a point is totally disconnected) and open. (A function which is continuous,
light and open is called an interior mapping. It is a famous theorem of StoIlov
that an interior mapping f of a plane domain is of the form f = oh, where
h is homeomorphic and analytic.)
2.4. Covering
Groups
Let S be a surface and (W,f) its smooth covering surface. A cover transformation g of W over S is a homeomorphism g: W W such that fog = f.
All such mappings g form a group G which is called the covering group of W
over S.
Two points of W equivalent under G have the same projection on S. 1f

conversely, any two points lying over the same point of S are equivalent
under G, then G is said to be transitive.
Let us consider the quotient space W/G and furnish it with the quotient
topology. Under certain conditions, WIG is a surface which is homeomorphic
to the surface S.
Theorem 2.2. If the projection mapping f: W S is sur)ective and the covering
group G of WDver S is transitive, then WIG and S are homeomorphic.
write [p] e WIG for the equivalence class containing the point
pEWand prove that
PROOF. We
139
[p]
f(p)
(2.2)
a homeomorpbism of WIG onto S. First, it follows from f = fog, g E G,

that (2.2) is well defined in WIG. It is surjective, because f: W S is onto, and
is
injective, because G is transitive. Its continuity follows from the continuity of

f: W .
S,
and the continuity of its inverse from the fact that f: W . S
locally homeomorphic.
is
The points of W are isolated with respect to 0 in the following sense:

Each point of the surface W has a neighborhood in which no two points are
equivalent under the action of the covering group.
In fact, it follows from the definition of a cover transformation that an

open set in which the projection mapping f: W , S is injective cannot contain
points equivalent modulo G. From this observation we can draw another

conclusion;
Except for the identity mappiFig, a cover transformation has no fixed points.
For assume that q is a fixed point for a transformation g 0. Since g is

continuous, it maps a point p near q onto a point g(p) near
= q. Because
p and g(p) are equivalent under G, we conclude that for all p in a sufficiently
small neighborhood of q, we have g(p) = p. It follows that the set in which
g(p) p is open. It is also closed and nonvoid. Since a surface is connected,
we see that g is the identity mapping.
For the most part, we shall be dealing with the covering groups corresponding to universal covering surfaces.
Theorem 2.3. The covering group of a universal covering surface W over
surface S is transitive.
Suppose that a and a' are points of W which lie over the same point
of S. Choose a point p W, join a to p by a path on W, project this path onto
PROoF.
5, and lift the projection back, but from the point a'. Let p' be the terminal.
point of this lift. We define g by the condition g(p) = p', and check that g is
well defined and a cover transformation of W over S. Hence a and a' are
equivalent under the covering group.
0
Combined with Theorem 2.3, Theorem 2.2 says that for a universal covering
surface W of S, the space WIG is always homeomorphic to S. The following
result sheds addinonal light on this connection.
Theorem 2.4. The covering group of a universal covering surface of S is isomorphic to the fundamental group of S.
140
PRoOF. Given a point a E W, let y be a closed path on S from f(a), and be W

the terminal point of the lift of y from a. Then a and b both lie over f(a). By
Theorem .2.3, there is a unique cover transformation with
property
g7(a) = b. It is easy to verify that [y] , is the desired group isomorphism
(cf. AhlforsSario (1], p. 38).
Li
2.5. Properly Discontinuous Groups

Starting with a given surface S, we arrived via a covering surface (W,f) of S
at the covering group G of W over S. Theorem 2.2 tells that, under very
general conditions, the circle from S to W to G closes, in the sense that the
quotient W/G is homeomorphic to S.
We shall now take a different starting point and prescribe directly a surface
W together with a group G of homeomorphic self-mappings of W. Again, we
form the quotient space WIG, furnish it with the quotient topology, and want
to impose a condition on G which makes WIG a surface.
For a covering group, every point has a neighborhood in which no two

points are equivalent. However, this property does not characterize covering
groups. In fact, examples can be given of groups G which possess this property but for which WIG is not even a .Uausdotff space. We need a stronger
condition on G.
A group G acting on W is said to be properly discontinuous if for any two
compact sets A, B W,, the intersection g(A) B is void, except for finitely
many g e G. Unlike a covering group, a properly discontinuous group need
not be fixed point free.
A point pe W is a limit point of a group G acting on W if there are distinct
mappings g,, e G, n =
1,
2,
..., such that p = urn
for some point q e W. A
properly discontinuous group has no limit points. This follows immediately

from the definition of proper discontinuity.
A fixed point free properly discontinuous group G shares the property of
covering groups that every point pe W has a neighborhood in which no two
points are equivalent modulo G. For assume that there are two sequences of
different points
a compact neighborhood A of p such that
lim a,, = urn b,, = p and b,, =
g,, e G. Then g,,(A) n A 0.
If there are infinitely many different mappings g,,, then G is not properly
discontinuous. If there are only finitely many different transformations g,,.
then at least one of them appears infinitely many times in the sequence. For
such a mapping p is a fixed point.
Theorem 2.5. A transitive covering group is properly discontinuous.
PROOF. Let (W,f) be a covering surface of S, and suppose that the covering
group G of W over S is transitive. Given two compact sets A and B in W, we
141
consider the subset C =

= f(q)} of A x B. The complement of C
is open, because f is continuous and two different points f(p) and f(q) have
disjoint open neighborhoods in S. It follows that C is closed and hence

compact.
For g e G fixed, we write U9 = {(p, q) e Cl q = g(p)}. Each U, is open in C,
because! is a local homeomorphism. By the transitivity of G, the union of the
disjoint sets U,, where g runs through all elements of G, agrees with C. Since
C is compact, We conclude that only finitely many of the sets U, are nonempty. Consequently, G is properly discontinuous.
U
The following result is a converse to Theorem 2.2.
Theorem 2.6. Let W be a surface, G a properly discontinuous fixed point free
group of homeomorphisms of W onto itself, and f: W -+ WIG the canonical
projection. Then
1. W/G is a surface,
2. (W,f) is an unlimited covering surface of WIG,
3. G is the (transitive) covering group of Wover WIG.
PROOF. By definition,f is continuous. If A W, then f'(f(A)) = ug(A),

g e G, from which we conclude that f is open.
In order to prove that WIG is a Hausdorif space we consider two different
points f(a) and f(b) of W/G. Since G is properly discontinuous, there exists a
compact neighborhood B of b which does not contain any point g(a), g E G.
After this we conclude the existence of a compact neighborhood A of a

such that A g(B) is empty for every ge G. Then g1(A) g2(B) = 0 for all
g1, g2 e G, and it follows that f(A) and f(B) are disjoint neighborhoods of
f(a) and f(b).
Clearly WIG is connected and has a countable base for topology. In order
to find local parameters we fix a point p e W. Since G is properly discontinuous and fixed point free, there exists an open neighborhood U of p such
that g(U) U = 0 for all mappings g G different from the identity. Then
fl U is injective, and if U is so small that it lies in the domain of a local
parameter h of W, then ho(fIU)' maps the open set f(U) in WIG homeomorphically onto an open set in the plane. Since f: W W/G is surjective, it
follows that WIG is a surface. Also, (W,f) is a smooth covering surface of
WIG.
From the definition it is clear that every g G is a cover transformation.

Conversely, let w be a cover transformation and p E W. Then there is a g E G
such that g(p) = w(p), for otherwise we would have f(w(p)) f(p). Hence
w =9.
Since G is a transitive covering group, it is not difficult to show that (W,f)
is an unlimited covering surface of WIG (cf. AhlforsSario [1], p. 29).
0
142

3.1. Lifted and Projected Conformal Structures
Let us now apply the results of section 2 to Riemann surfaces.
Theorem 3.1. Let S be a Riemann surface and (W,f) a smooth covering surface
of S. Then Wcprries a unique conformal structure which makes the projection
mapping fanalytic.i
PROOF. Let H be the conformal structure of S. For every point p e W we
choose a neighborhood U of p such that fI U is injective and f(U) is contained in the domain of some heH. Then the atlas {ho(ftU)Ipe W} defines
a confonnal structure for W, and f is analytic with respect to this structure.
We say that this conformal structure of W is obtained by lifting the

conformal structure of S. If the projection f: W S is analytic with respect
to a conformal structure of W, then the condition which expresses this fact
shows directly that this structure is the same as the lifted structure. Thus the
uniqueness assertion in the theorem follows.
El
Using the characterization (2.1) of a covering surface, we could show

without difficulty that Theorem 3.! holds also in the case where (W,f) is a
branched covering surface of S (cf. AhiforsSano J1J, p. 119).
In the sequel, a covering surface of a Riemann surface is always regarded
as the Riemann surface with the lifted conformal structure.
The following observation is immediate: Let S be a R'temann surface and
(W, f) a smooth covering surface of S. Then the cover transformations of W

over S are conformal. For locally a cover transformation g is of the form
(flg(U)Y1 oft U, and hence conformal.
Theorem 2.6
be refined in the setting of Riemann surfaces.
Theorem 3.2. Let W be a Riemann surface, G a properly discontinuous fixed

point free group of conformal self-mappings of W, and f: W -+ WIG the canonical projection. Then the surface WIG carries a unique conformal structure which
to the original conformal structure of W.
This follows immediately from the way the local parameters of WIG were
defined in the proof of Theorem 2.6. In the situation of Theorem 3.2, the
conformal structure of W is said to have been projected to WIG. If W is a
given Riemann surface, we always regard the quotient WIG as the Riemann
surface with the projected structure.
Suppose that G is a properly discontinuous group of conformal selfmappings of a Riemann surface W, bin not fixed point free. We can still
143
conclude that WIG is a surface and that WIG carries a conformal structure
which makes the projection mapping f: W * WIG analytic. However, W is
now a covering surface of WIG which is branched at the fixed points of G
(AhlforsSario [1], p. 121).
3.2. Ricmann Mapping Theorem

Our results in 2.4, 2.5 and 3.1 lead to a fundamental representation of
Riemann surfaces if they are combined with the following generalization of
the Riemann mapping theorem for plane domains.
Theorem 33 (Riemann Mapping Theorem). Every simply connected Riemann
surface is conformally equivalent to one and on(y one of the following plane
domains: the unit disc, the complex plane, or the extended plane.
lengthy preparations.
This is a deep result, and a complete proof
We content ourselves, therefore, with sketching the main lines of a proof
based on the use of subharmonic functions. (Subharmonic functions are
defined on Riemann surfaces with the aid of
parameters. This is possible
because subharmonicity is a local and conformally invariant property.)
First of all, a classification of Riemann surfaces into compact. parabolic,
and hyperbolic surfaces is needed. A non-compact Riemann surface S is

parabolic if every negative subharmonic function on S is constant; otherwise
S is hyperbolic.
Using subharmonic functions and Perron families, we can define Green's
functions for Riemann surfaces just as it is done for the case of plane domains.
The Green's function of a Riemann surface S with singularity at the point
peS isa function which is positive and harmonic on S (p}. To describe its
singularity, we consider a local parameter z mapping a neighborhood of p
onto the unit disc such that z(p) = 0. Then it is required that + logizi be
harmonic at p; this is an invariant definition not depending on the choice qf
the local parameter. The Green's function is characterized by the property
that among all functions positive and harmonic on S {p} and possessing
the same singularity at p as gp, the function 9,, is the smallest. If a Green's
functioa exists for some p ES, then it exists for every p eS. By a theorem of
Ohtsuka, the Green's function exists if and only if S is hyperbolic.
If S is parabolic or c.ompact, Green's functions do not exist but it is

possible to prove the existence of a function Up,q with the following properties:
is harmonic in S {p} {q}; if z(p) 0, then
is
harmonic at p, and if z(q) = 0, then u,,q +
is harmonic at q; outside
parameter discs (preimages of discs under z) containing p and q, the function
is bounded (Nevanlinna [1], p. 212).
Suppose now that S is simply connected. If S is hyperbolic, we take a

in a parametric disc, and extend
Green's function g,,, form its conjugate
144
+ ig)) by analytic continuation to S. Using the Monodromy theorem, we conclude that the extended function is single-valued on S. Finally,
application of the maximum principle shows that it maps S conformally onto
the unit disc (Nevanlinna [lJ, p. 204).
If S is parabolic or compact, a conformal mapping of S into the extended
plane can be constructed with the aid of the function Up,q (Nevanlinna [1],
p. 213). In case S is parabolic the boundary of the image consists of one point,
whereas the boundary is empty if S is compact.
Theorem 3.3 can also be proved by a method which is based on the use of
quasiconformal mappings (Bers [5]). The idea is to first construct a topological and locally quasiconformal mapping of the given Ricmann surface S
into the plane, then apply the existence theorem of Beltrami equations to
obtain a conformal mapping of S .(as Gauss did; cf. 1.6), and complete the
proof with the aid of the Riemann mapping theorem for plane domains.
Theorem 3.3 occupies a central position in the theory of Riemann surfaces.
It is often called the general uniformizatioi theorem. The first proofs are
attributed to Koebe (in 1907) and Poincar.
exp(
3.3. Representation of Riemann Surfaces
Let S be an arbitrary Riemann surface and (W,f) its universal covering

surface. Since W is simply connected, we conclude from Riemann's mapping
theorem the existence of a conformal mapping w: W D, where D is the unit
disc, the finite plane or the extended plane. But then (D, Jo
is also a
universal covering surface of the Riemann surface S, and we have proved the
following important result:
Every Riemann surface admits as its universal covering surface the unit disc,
the finite plane, or the extended plane.
This makes possible a far-reaching normalization of universal covering

surfaces. A consequence of basic importance is the fact that the elements of
the covering group of D over S, being conformal self-mappings of D, are
Mbius transformations.
Summarizing the topological results in 2.45 and the analytical results in
3.12, we obtain the basic representation theorem for Riemann surfaces.
Theorem 3.4. Given an arbitrary Riemann surface S. let D be its universal
covering surface, and G the covering group of D over S. Then S is conformally
equivalent to the Riemann surface DIG.
PROOF.
It follows from Theorems 2.3, 2.5, and 3.2 that the quotient DIG is a
Riemann surface with the projected conformal structure. By Theorem 2.2, the
mapping (2.2) is a homeomorphism of DIG onto S. It is conformal, because
the conformal structure of S is also obtained by projection from D.
0
145
Theorem 3.4 is the analytic counterpart of the topological result expressed

in 2.4 that every surface S is topologically equivalent to the quotient WIG,
where W is a universal covering surface of S.
Theorem 3.4 can be supplemented as follows: Let G be an arbitrary

properly discontinuous group of conformal mappings acting on D. Then DIG
is a Riemann surface. This follows from Theorem 3.2 and the remark made
after it.
3.4. Lifting of Continuous Mappings

We have seen above that, under certain circumstances, a mapping between
universal covering surfaces projects to a mapping between the underlying
surfaces. Here we take the opposite view and show how to lift mappings
between Riemann surfaces to mappings between their universal covering
surfaces.
Let q
a continuous mapping of a Riemann surface
into another
universal covering surface of
I = 1, 2.
D2. By 3.3, we can choose this common
surface, which we denote by 0, so that D is the unit disc, the complex plane
or the extended plane.
be
Riemann surface S2, and (D1, it1)

Here we consider only the case
In the extended plane every Mbius transformation has a fixed point.

Thus, if the extended plane is the universal covering surface of a Riemann
surface, the covering group over this surface is trivial. By Theorem 3.4, the
surface itself is then the extended plane up to conformal equivalence. Therefore, we can exclude this trivial case in what follows.
The continuous mapping 0: S1 . S2 always induces a continuous mapping
of D into itself. More precisely, there is a continuous f: D ' D such that
=ir2of.
(3.1)
The construction off, which is called a lift

is as follows: Fix first z0eD
and w0
{p(1t1(z0))}. If y is a path in 0 from z0 to:, we define f(z) as the
endpoint of the path which we obtain by lifting po it1 oy from w0. If y' is
'!nother path in D from z0 to z, then q' o o y and qo
it follows from the Monodromy theorem that f(z) is well defined. From
the definition it is clear that f is continuous.
The mapping q induces a mapping of the covering group G1 of D over S1
into the covering group 62 of D over S2: We shall show that the relation
O(g)of = fog
(3.2)
defines a mapping 0 which is a homomorphism of G1 into G2.
in order to prove that (3.2) defines a homomorphism 0: G1
G2, we
choose an element geG1. From (3.1) it follows that

For every z D we thus have an element h
G2
such that f(g(z)) = h(f(z)).
__jw;
LV. Riemann Surfaces
146
0
0(g)=fgf'
>u-
>s2
Figure S. Mappings induced by a homeomorphic
The mapping h depends only on g, not on the point z. For if y is a path in D

from z to z', then fogoy and hofoy start at the same point and have the
op. It follows that they have the same terminal point,
same projection
i.e., the element of 62 corresponding to z' is also h. Hence (3.2) holds for
0(g) = h. From
= 0(g2og1)of.
O(92)ofogl
we see that the mapping 0 is a homomorphism.

The mappings f and 0 induced by -p are not unique. 1ff is a lift of
then
fog is also a lift of for every geG1. From fog = 0(g)of it follows that
such lifts are of the form hof, where h = 8(g) is an element of 02. The
mappings
hof,
heG2,
represent all possible lifts of q. We see this by repeating the reasoning which
showed that (3.2) defines an element 0(g) of 62.
1ff induces the group homomorphism 0, then hof induces the homomorphism g i ho 0(g)o 1r1, which differs from 0 by an inner
of G2. We call two such homomorphisms equivalent and conclude that all
homomorphisms induced by qi: S1 S2 are equivalent.
S1
S2 and o
S1
S2 determine equivalent homoSuppose that
induces 0, then
can be so lifted that it also induces 8. In
morphisms. If
induces a homomorphism g ho 0(g)o h1, he G2, and so
fact, a lift f1 of
h'of1, which is a lift of induces 0.
If q: S1 + s2 is a homeomorphism, then so is every lift f: D -+ D. in this
onto G2. (Fig. 8.)
case g ' 0(g) = fog of is an isomorphism of
3.5. Homotopic*4appings
Lifting of mappings is closely related to the topological notion of homotopy.
t l} the unit interval.
S2 be a homeomorphism and I
Let 600: S1
{tIO
A homeomorphism p S1
if there is a
S2 is said to be horn otopic to
147
continuous mapping h: S1 x
such
I p
mapping h is called a homotopy from
that h(.,O) =
h(., 1)
The
to
the
Given a lift J0 of
Suppose that h is a homotopy from
to
1, can then be so lifted that we obtain a
mappings h(.,t): S1 . S2, 0 t
This homotopy lifting property follows
homotopy from to a lift of
easily from the definitions.
Theorem 3.5. Two homeomorphisms

group isomorphisms
PROOF.
to
and only
q1: S1 ' S2.
I=
0,
1, induce the same
they are homotopic.
is homotopic to qi1. Let h be a homotopy from

and f, a lift of h(., t), 0 t
1, such that 1, is a homotopy between
Assume first that
f0andf1.
Choose g G1 and z eD, and consider the two paths t '
and
) (J(z)). Both have the same initial point f0(g(z)) and the same
projection t *
on S2. Hence they agree, and for t
I we obtain the
t *
(f0 o g
desired result
(3.3)
Assume, conversely, that p0 and
have lifts .f0 and f1 such that (3.3)
holds for every geG1. If D is the unit disc, we define
0 < t < 1, as
follows: f,(z) is the point of the hyperbolic geodesic arc joining f0(z) and .11(z)
which divides the hyperbolic length of this arc in the ratio t: (1 t). Then J
is a homotopy between f0 and f1.

Under the mapping 0(g) = o go
(= o g ofr') the endpoints of the
arc map to f0(g(z)) and f1(g(z)). But since 0(g) leaves hyperbolic distances
invariant, 0(g) maps the point f,(z) to
Hence, 9(g) of; = f; og. In other
words, all mappings 0 t 1, determine the same group homomorphism.
it
It follows that
and
z + b.
is the finite plane, all cover transformations are translations z
Therefore, the above reasoning remains valid if the' hyperbolic metric is

replaced by the euclidean.
3.6. Lifting of Differentials
Let S and W be Riemann surfaces and f: W - S a non-constant analytic

mapping. It follows from the definition of a covering surface in 2.3 that (W,f)
is a covering surface of S (ci. Ahlfors-Sario [1], p. 119).
Let the conformal structures of W and S be detennined by the local

parameters and Ic1, respectively. Given an arbitrary point of W, we consider
an open neighborhood V of this point which is contained in the domaiti of a
local parameter h1 and is so small that f( V) lies in the domain of a local
parameter k1. For p e V, we write z1 = h,(p), =
Let be an (m, n)-differential on S (ci. 1.4), and 'let denote its represen-
148
tation in the local coordinate wj
Set
=
(3.4)
By formula (1.2), the function element

does not depend on the particular choice of the parameter made in 1(V). We now require that (3.4) remains valid when we change the parameter of W in V. Then (3.4) defines an
(m, n)-differential on V, and since we started with an arbitrary point of W, on
the whole surface W. It is called the 4ft of to W.
We can say a little more: Formula (3.4) shows that f: W ' S induces a
linear mapping of the space of holomorphic (m, n)-differentials of S into the
space of holomorphic (m, n)-differentials of W. 1ff is a conformal mapping of
W onto S, the induced mapping is bijective.
Suppose, in particular, that W = D is the universal covering surface of 5,
where D is the unit disc or the complex plane. In both crises the conformal
structure of S is determined by the local inverses of the projection mapping
of D onto S. The conformal structure of D is of course given by the identity
mapping z z.
What we gain from the use of the universal covering surface D is that we
now possess a global coordinate z eD for the representation of (p. In other
and formula (1.2) shows
words, in (3.4) we can put z, =
= z. Thus
that every is the restriction of a function globally defined in D. We denote
this function by p, i.e., we identify it with the collection of its restrictions. It
follows that the !!ft of the differential q, of S to the universal overing surface D
is a global representation of q in terms of the coordinate z e D.
Let g be an arbitrary element of the covering group G of D over S. Suppose

z is a local parameter in an open subset U of S defined by the inverse of
a suitable restriction of the projection mapping. Then p , g(z) is a local

parameter in U such that z and g(z) have the same preimage in U. Hence, it
follows from the invariance (1.2) that
p(z)
(3.5)
for every geG.

Conversely, if is a complex-valued function in D with the property (3.5),
then defines an (m, n)-differential of S. A function (p satisfying (3.5) is said to
the group G. Thus there is no difference whether we
be an (m,
interpret to be a differential on the Riemann surface S or for the covering

group G of D over S.
As one application, we introduce the hyperbolic metric, which we have
so far considered in plane domains conformally equivalent to a disc, to

Riemann surfaces. Let S be a Riemann surface which has the unit disc D as
its universal covering surface. The Poincar density z -.
= 1/(1 1z12) of
0 satisfies the condition
(ng)lg'I =
(3.6)
4. Groups of Mbius Transformations
149
for every cover transformation g. It follows that is a (1/2, 1/2)-differential on

S, and
= 11(1 1z12) is its global representation in the coordinate z ofD.
Now let y be a rectifiable arc on S. We define its hyperbolic length to be
equal to the hyperbolic length of its lift to D. Because of the invariance (3.6)
it does not matter how we choose the point of D lying over the initial point
of y.
In order to study the geodesics on 5, we take two points p and q of S and

join them by an arc y. Let z0 be a point of D over p. and let z1 denote the
terminal point of the lift of y from z0. Then the projection of the hyperbolic
geodesic from z0 to z has the shortest hyperbolic length in the homotopy
class of y. The infimum of these shortest lengths over all homotopy classes is
the hyperbolic distance between p and q. The infimum is attained. In fact, the
hyperbolic geodesic in D joining z0 to a "closest" point of the preimage of q
projects to a geodesic between p and q.

4.1. Covering Groups Acting on the Plane
Let S be an arbitrary Riemann surface. We again normalize its universal
covering surface D so that D is the unit disc, the complex plane or the
extended plane, and denote by G the covering group of D over S. In view of
the representation S = DIG modulo conformal equivalence, the theory of
Riemann surfaces can be regarded as essentially equivalent with the theory of
discontinuous groups of Mbius transformations acting on D. Lehner ([1],
Chapter 1) gives an interesting survey of the historical development of the
theory of Mbius groups.
The points of DIG are called orbits of G. A subdomain of D is said to be a
fundamental domain of G if it contains at most one point of every orbit of G
and its closure in D meets evi..ry orbit of G.
In the cases where the universal covering surface D is the extended plane
or the complex plane, all possible covering groups of D over S can be readily
listed. We know already that if D is the extended plane, the covering group G
is trivial and S is conformally equivalent to D.
Suppose next that D is the complex plane. Since the elements of G have
their fixed point at
they are translations z -. z + a. Here three possible
types of discontinuous groups G arise. First, G may be trivial, in which case
S is conformally equivalent to the finite plane. Second, G can be infinite,
generated by a transformation z -+ z + w, w 0. A fundamental domain of
such a group is the interior of the parallel strip bounded by straight lines
through 0 and through w and perpendicular to the vector from 0 to
w.
Topologically, D/.G is an infinite cylinder. The function z . exp(2miz/w),
150
IV.
Surfaces
which is invariant under G, shows that DIG is conformally equivalent to the
finite plane punctured at 0.

A third possibility is that G has two generators z ' z + w1, z z + q2,
0. A fundamental domain is now the interior of the parallelo+ co2, and a2. In this case the Riemann
gram P with vertices at 0,
surface S = DIG is compact. For the closure P is compact and S is the image
of P under the continuous projection mapping D DIG.
the opposite sides of P are equivalent under G, it follows that

topologically S is a torus. Two different tori obtained in this fashion are
Since
generally not conformally equivalent. The conformal structures on a torus

will be studied in V.6.
A simple geometric argument shows that there are no other ways to form
groups of translations which are properly discontinuous in the finite plane.
4.2. Fuchsian Groups

Let us now consider the case in which the Riemann surface S admits a disc D
as its universal covering surface. It follows from the results of section 3 that
the covering group G of D over S is a properly discontinuous fixed point free
group of Mbius transformations which keeps the disc D invariant. We

call such groups Fuchsian groups. (In the literature, fixed points are usually
allowed.) Conversely, every Fuchsian group G acting on D is the covering
group of D over the Riemann surface DIG.
An arbitrary Mbius transformation z ' w with two finite fixed points z1
and z2 has the representation
wz1
WZ2
= pet
If z2 =
= pe'(z zr). The geometric action of a Mbius

we have w
transformation is best seen from this representation, which also gives rise to
the division of Mbius transformations (different from the identity) into four
classes. If p = 1, 0 0, the transformation is elliptic, if p 1, 9 = 0, it is
hyperbolic, and if p # 1, 0 0, it is loxodromic (0 0 < 2ir). A Mbius
transformation with only one fixed point is parabolic. The class of a Mbius
transformation g remains the same when g is conjugated by an arbitrary

Mbius transformation h, i.e., when g is replaced by hogoh1.
A loxodromic transformation does not keep any disc invariant. If an
elliptic transformation g maps a disc D onto itself, then one of the fixed points
of g lies inside D while the other is its mirror image with respect to 3D.
Let a Fuchsian group act on a disc D. Since it has no fixed points in D, it
follows from the reflection principle that the elements of the group do not
have fixed points in the complement of the closure of D either (with the
identity mapping excluded of course). We conclude that the elements of a
Fuchsian group are hyperbolic or parabolic, with fixed points on
151
In studying a Fuchsian group G, we choose the invariant disc on which G
acts to be the upper half-plane. The elements of G are then of the form
z --s g(z)
az + b
+ d'
where the coefficients a, b, c, d are real and ad bc > 0. Groups of Mbius

transformations with real coefficients are said to be real.
The conformally invariant hyperbolic metric ds = IdzI/2 Imz of H is a

natural tool for the study of the geometric properties of U. As before, we use
for the distance between the points z1 and z2 of H in
the notation
this metric. We fix a point z0H and denote by G(z0) the orbit of z0. For
every
= 0, 1,... ,we write
=
for all
are non-empty, open and mutually disjoint, and the union of the
The sets
=
closures in H of all
It follows
is H. If Z,, = g(zj) for g G, then
that all sets N, are congruent in the non-euclidean geometry of H. In studying
we may therefore restrict ourselves to one of

the properties of the sets
them, say to N0, for which we also use the shorter notation N.
A point zeH lies on the boundary of N if and only if h(z,z0) <
for all Zt e G(z0) and equality holds for at least one Zk, k 0. The set
{ZEHIh(z,:0) = h(z,zk)} is the non-euclidean line which is the perpendicular
bisector of the non-euclidean segment joining z0 and Zk. It follows that N is
a convex polygon; in particular, N is connected. N is called the Dirichiet
region of U with center at z0.
From the definition of N we conclude that N is a fundamental domain of

G. Its boundary arcs lie either on the real axis, in which case they are said
to be free sides, or they are situated in H, with the possible exception of
endpoints on Il', and are called inner sides of N.
The inner sides of N are pairwise equivalent under G, whereas inner points
of a free side have no equivalent points in the closure of N. These properties
of N can be deduced without difficulty from the definition. More careful
analysis is required to prove the following fundamental result:
Theorem 4.1. The elements of a Fuchsian group which map the inner sides of a
Dirichlet region onto each other generate the whole group.
Dirichlet regions are studied in detail in Springer [1]; for the proof of
Theorem 4.1, see p. 237.
4.3. Elementary Groups

A group of Mbius transformations can of course be regarded as acting on
the extended plane. We shall now adopt this point of view.
The limit set L of a group G of Mbius transformations is the set of the
152
limit points ofG. From the definition of a limit point, which was given in 2.5,
it follows that g(L) = L for every g e G. Also, L is closed (cf. Lehner [I], p. 88).
If L contains at most two points, the group G is said to be elementary.
MI elementary groups can be listed. (A comprehensive treatment of elementary groups is given in Ford [1], Chapter VI.) First of all, a finite group
is necessarily elementary, its limit set being empty. If the Riemann sphere
is rotated so that a regular solid remains invariant, then stereographic pro-
jection leads to Mbius transformations which form a finite group. All

non-cyclic finite groups of Mbius transformations are obtained from such
groups by conjugation.
The elements of a finite group are elliptic transformations. They are of
finite period, i.e., there is a natural number n such that the nth iterate of the
transformation is the identity mapping. A cyclic group generated by an

elliptic transformation which is not of finite period is very different: Every
point of the plane is a limit point of such a group(Lehner [1], p.87).
There is a second type of elementary groups G all of whose elements are
elliptic or parabolic transformations sharing a common fixed point. Then G
has this common fixed point as its sole limit point (Lehner [1], p. 93). A
simple example is the group generated by the elliptic transformation z z
and the parabolic transformation z -+ 2 + 1. In this case L =
J.
Any other infinite group is elementary if and only if it is cyclic and the
generator is not elliptic. The limit set L then agrees with the set of the fixed
points of the generator (Lehner [1), p. 87). Thus Lconsists of a single point if
the generator is parabolk and of two points if the generator is hyperbolic or
loxodromic.
An example of an elementary
group acting on the upper halfplane H is the real cyclic group
a>I.
In this case L = (0,
The function
z
(4.1)
is invariant under G. By studying the image of the fundamental domain

{z 6 HI I <Izi <a) under (4.1) we deduce that the annulus
= (WI! < wi <

a model of H/G. In other words, G is the covering group of the upper
A
is
half-plane H over the annulus A.
4.4. Kleinian Groups

Returning to an arbitrary group G of Mbius transformations, we denote by
the complement of the limit set L with respect to the plane. A point of is
called an ordinary point of G, and Q is said to be the set of discontinuity of G.
153
4. Groups of Mobius Transformations
The set (1 can be empty. A trivial example is the group of all Mbius transformations. But fl can be empty even for a cyclic group: We pointed out
in 4.3 that this is always the case if the group is generated by an elliptic
transformation which is not of finite period.
Since L is closed, fl is open. The set C need not be connected. From the
invariance of L under G it follows that g(Q) Cl for every g E G.
If Cl is not empty, G is called a Kleinian group. A Kleinian group is

countable (Lehner [1), p. 90). Fuchsian groups and elementary groups are of
course special cases of Kleinian groups.
Let g be a Mbius transformation and g(z) = (az + b)/(cz + d) its unimodular representation, i.e., ad bc = 1. All 2 x 2-matrices with determinant 1 form a group SL(2) under matrix multiplication. The mapping
g_+(
(4.2)
is an isomorphism of the group M of all Mbius transformations onto the

quotient group SL(2)/ 1, where I is the identity matrix.
If the distance of the matrices
and (ba) is defined by
max{Iajj
= 1,2),
then SL(2) becomes a topological group. Via the mapping (4.2), the topo-
logical structure is transferred to M. A subgroup G of M is called discrete if

its elements are isolated in the topology of M. It is not difficult to prove that
0 is discrete if and only if it does not contain infinitesimal transformations,
i.e., if and only if there is no sequence of distinct elements
G, n = 1, 2,...,
such that jim
= z for every z
[1], p. 96). From this characterization of discreteness we conclude that if G is not discrete, then every point of
the plane is a limit point of G. It follows that a Kleinian group is discrete. The
converse is not true.
The set of discontinuity Cl can be characterized by means of normal

families. Let A be a domain of the plane and G a Kleinian group. The family
{9IAIgeG} is normal (f and only 4f A is a subdomain of fl(Lehner [1], p. 98).
In particular, if Cl is connected, then Cl is the largest domain in which the
mappings g e G constitute a normal family.
4.5. Structure of the Limit Set

The normal family criterion for sets of discontinuity can be used to proving
the following result, which reveals several properties of the limit set (Lehner
[1], p. 103).
Lemma 4.1. For a Kleinian group G, euery point L is the cluster point of
each orbit G(z), with the possible exception of z = and one other point z EL.
IV. Riemann Surfacec
1 58
We first deduce from this lemma that if G is not elementary, every poinuof
L is the cluster point of other limit points. Hence, L is then always a perfect
set. It follows that for Mbius group:; there is a striking dichotQmy: Either the
limit set contains at most two points or else it contains uncountably many.
points.
A second conclusion from Lcmma 4.1 is that the limit set of a Kleinian
group agrees with the boundary of the set of discontinuity. For we have trivially
=
so that
L. On the other hand, we infer rom Lemma 4.1
that L c i. Hence, L c: Li n L =
and we obtain the desired result
L
(4.3)
Third, Lemma 4.1 (or (4.3)) shows that the limit set of a Kleinian group is
nowhere dense in the plane. For to every
there is a point z e and
For Fuchsian groups the same reasoning
mappings e G, such
gives the following result:
The limit set of a

boundary
group acting on a disc D is either the whole
or a nowhere dense subset of
If the limit set of a Fuchsian group G agrees with the boundary of the
invariant disc, G is said to be of the first kind (or horocyclic). Otherwise, G is
ti has
of the sccond kind. 41 follows that for groups of the first
components, whereas i is connected if the group is of the second kind.
For a Kicinian group G,let F denote the set of the fixed points of its
elements (other than the identity). If g e G and z is a fixed point of g0 E G, then
g(z) is fixed point of go Yo g
Hence g(F) = F for every g E G.
For the group generated by z z + 2 and z l/z, the set is the union
of the upper and lower half-planes while the point i belongs to F. This shows
that F need not be a subset of L even though the group is infinite. But
Kleinian group does not contain elliptic transformations of finite period, then
the closure ofthe set of its fixed points coincides with its limit set.
In particular, if G is a Fuchsian group of the first kind acting on the upper
half-plane, then the fixed points of G are everywhere dense on the real axis.
There are even sharper relations between F and L. Let Fh, F,, and F,,
denote the subsets of F consisting of the fixed points of the hyperbolic,
loxodromic and parabolic elements of the Kleinian group G. Then
L=F,,
L=P,,,
(4.4)
an element from the class iii question. The relations

(4.4) can be proved with the aid of Lemma 4.1 (cf. Lehner [1], p. 104).
Suppose that G is a Fuchsian group of the second kind acting on H. Then
is a bordered Riemann surface with

given in 1.3).
\ L)/G
as its border (cf. the definition
155
Groups of Mbius Transformations
From this representation of a bordered Riemann surface 5* we see that S*

can be imbedded in a larger Riemann surface. In fact, interpreting G as acting
on the plane, we form the quotient Q/G. By Theorem 3.2, it is a Riemann
S = H/G. The
surface, and clearly it contains
\ (H u R))/G is called the mirror image of S = H/G, no
Riemann surface
matter whether G be of the first or of the second kind.
4.6. Invariant Domains

The components of the set of discontinuity Q of a Kleinian group G are
disjoint domains. A component which is mapped onto itself by every element
of G is called an invariant domain of G.
A Kleinian group G with no fixed points in Q is Fuchsian if it has a disc in
0 as an invariant domain, and it is said to be quasi-F uchsian if it has a Jordan
domain in 0 as an invariant component.
The following result makes it possible to analyze invariant domains.
Lemma 4.2. Let G be a Kleinian group such that 0 has an invariant component
A which is a Jordan domain d!/Jerent from a disc. Then A does not have a
tangent at a fixed point of a loxOdromic element of G.
PROOF. Assume
that the tangent exists at a fixed point of a loxodromic
element g e G. We may suppose without loss of generality that the fixed point
of g lies at z = 0, that the tangent at z = 0 is the real axis and that cc is the
repulsive fixed point of g. Then g(z) =
where 0 < r < I and 0 < 0 <2it.
Suppose first that (I
it, and set
a = min(0/2,Iir
then 0 <a n/4. Consider the two angles
0)12);
or oe(ir a,
=
Since the real axis is a tangent, we have for every a > 0 a disc
D0 centered at the origin, such that
it
a). p
(4.5)
Now choose zeM
z
On the other hand,
0. Then
it follows from the definition of a that g(z)
This contradicts (4.5).
If z . g(z) = rz belongs to G, then go g is a hyperbolic transformation
with the same fixed points as g. A modification of the above proof shows that
3A does not have a tangent at a fixed point of a hyperbolic element of G. This
proves the lemma.
Combined with our previous results on Kleinian groups, Lemma 4.2 yields
the following result.
156
Theorem 4.2. The boundary of an invariant component of a quasi-Fuchsian

group is either a circle or a Jordan curve which fails to have a tangent on an
everywhere dense set.
PROOF.. First, if A denotes an invariant component, we clearly have
From (4.3) we then conclude that L. If the group is not FuGhsian,
it
always contains loxodromic elements (Lehner [1], p. 107). By (4.4), we have
in this case 3A F,. Hence, the theorem follows from Lemma 4.2.
0
It was Klein who first noticed that such weiid invariant Jordan domains
exist. He obtained such domains by direct construction. For details, interesting pictures and almost philosophical comments on this unexpected
phenomenon we refer to FrickeKlein [1]. Here we shall only briefly explain
Klein's method. In V.3.4 we shall arrive at invariant 4uasidiscs in a completely
different manner.
Let {D,Ii = 1,2,. . } be a closed chain of discs, i.e., the discs D, are disjoint
but for eveiy i, the closure of the union UI)j,J i, is connected. We form the
group 6 whose elements are compositions of an even number of reflections
in the circles aDs. The elements of G are Mbius transformations. We say that
G is generated by the chain {D, }.
If the number of the discs in the chain is one, G is trivial, if it is two, 6 is
cyclic, and if it is three, 6 is Fuchsian. In case the complement of the closure
of the union UD, of all discs is not empty, it is easy to see that this complement is contained in the set of discontinuity fl of G. In other words, G.isthen
always a Kleinian group.
.
The importance of KIem's method derives from the fact that under very
general conditions,
has two invariant components which are complementary Jordan domains. This is the case, for instance, if the chain has only
a finite number (>2) of discs. The boundary of the invariant domain passes
through the points at which the closed discs touch each other and through
their reflected images (Fig. 9).
Figure 9. Klein's method of generating invariant domains.
5. Compact Riemann Surfaces
5.
157
Compact Riemaun Surfaces
5.1. Covering Groups over Compact Surfaces

Let G be a Fuchsian group acting on the upper half-plane H. We recall that
all Dirichiet regions of G are congruent in the non-eucidean geometry induced
by the hyperbolic metric on H. The Dirichiet regions clearly have a bounded
hyperbolic diameter if and only if their closures lie in H. This property

characterizes compact Riemann surfaces.
Theorem 5.1. Let S be a Riemann surface and G the covering group of the upper
half-plane H over S. Then S is compact Vand only the Dirichlet regions of G
are bounded in the hyperbolic metric of H.
PROOF. Suppose first that S is compact. Let N be a Dirichiet region with

center a. We consider the hyperbolic discs
= (zlh(z, a) < n}, n 1, 2
Their projections on S = H/G form an open covering of S. Since S is compact,
there is an n such that the projection of D, alone covers S. In other words, for
every z H there exists a mapping g e G for which h(g(z), a) <n. Now if zeN,
then h(z,a) h(g(z),a) for every geG. It follows that N c
Assume, conversely, that the closure of a Dirichiet region of G lies in H.
Then S is the image of a compact set under a continuous mapping and hence
compact.
Theorem 5.1 admits interesting conclusions.
Theorem 5.2. The covering group of the upper half-plane over a compact
Riemann surface is finitely generated and of the first kind.
PROOF. Let S be a compact Riemann surface and G the covering group of H

over S. The vertices of a Dirichict region of G cannot have a limit point in H.
Hence, by Theorem 5.1, a Dirichiet region for G has a finite number of sides.
We conclude using Theorem 4.1 that G is finitely generated.
In order to determine the limit set L of G, we consider an arbitrary point
x of the real axis and set U =
xl <r}. The hyperbolic distance
as y-O.
from the point x + iyeU to the semicircle = r tends to
On the other hand, by Theorem 5.1 the Dirichiet region containing x + iy

has a uniformly bounded hyperbolic diameter for every y > 0. It follows that
U contains a Dirichiet region for every r > 0. Consequently, x eL, and so L
is the whole real axis.
IV. Rientaun Surfaces
158
5.2. Genus
of a Compact Surface
Let S be a compact Riemann surface and 6 the covering group of Ii over S.

A Dirichiet region N for G is then a non-euclidean polygon with finitely
many sides, which are pairwise equivalent under G. Now it is possible to
transform N to another non-euclidean polygon which is also a fundamental
domain of G and whose sides follow each other according to the pattern
(5.1)
the sides a and li, being equivalent to and b, respectively. The number p is
at least 2. Such a polygon is called a normal polygon for C. The transformation
of a Dirichiet region to a normal polygon is described in Nevanlinna [I],

pp. 229230.
The pattern (5.1), which generalizes the pattern a1 b1 a'1 b'1 of a torus (see
4.1), can be regarded as a representation of the compact Riemann surface S.
But it is in fact of a more general character. Let S be an arbitrary compact
orientable surface. A fixed (itecessarily finite) triangulation of S leads in a
natural manner to a polygon representing 5, in which identified sides either
have the simple pattern a1a'1 or else the pattern (5.1), with p =
1,
2,
(Springer [ii p. 117). Here p does not depend on the triangulation by way of
which we arrived at it. The pattern a1 a'1 occurs if and only if S is a topological
sphere.
The number pin (5.1) is called the genus of the compact surface S. A sphere
is said to have the genus p = 0. It follows that a torus has the genus p = I and
every compact Riemann surface which has the upper half-plane as a universal
covering surface has a genus p> 1.

The genus characterizes the topology: Two orient able compact surfaces 're
hoineomorphic if and only if they have the same genus. In analogy with the case
of a torus, the topological type of a surface can be read from the pattern (5.1):
4
)mpact orientable surface of genus p is a topological sphere with p handles.

These two results are proved in Springer [11, which contains a detailed
discussion of the relations of genus to such topological invariants as the

fundamental group, homology groups, and the Euler characteristic. We shall
study here certain analytic implications of genus.
5.3. Function Theory on Compact Riemann Surfaces

Let f be a non-constant meromorphic function on a compact Riemann surface
S of genus p. lIp = 0, the surface S can be identified with the extended plane,
takes on every complex value
and / is a rational function. In
and cx, each of them the same number of times, provided of course that
multiple values are counted according to their multiplicities.
I. the study of f amounts to function theory on a torus. (Its eleIf p
159
5;'Compact.Rieharnt Surfaces
ments are described in Springer [1], P. 34.) The lift off to the universal covering surface C IS double-periodic, i.e., an elliptic function.
Regardless of p. the property of rational fitnctions remains trtk as indicated
above(Springer [1], p. 176):
.
Theorem 5.3. On a cOmpact Riemaun surface, a non-constant meromorphic

function assumes every value
same finite
of times.
For topological reasons, injective meromorphic functions vah exist only if
p = 0. The theory of meromorphic functions on compact Riemann sdrfaces is

classical analysis, which is intimately connected-wi th tile theory or
functions (see, e.g., Springer [1], p. 286). In a way, the theory was born beMre
the notion of a Riemann surface existed in afly form.
A striking example of the interrction between topology and analysis on

compact Riemann surfaces is provided -by the complex vector space consisting of holomorphic Abelian differentials. The study of the periods of
regular harmonic differentials on a homology basis yields the following result
(Springer [1], p. 252):
On a compact Riernann surface of genus p. the dimension of the space of

holoinorphic Abe/ian differentials
equal to p.
We shall use this result in determining the dimension of the complex

vector space formed by holomorpiiic quadratic differentials. This linear space
will play an important role in the theory of Tichrnller spaces.
-
5.4.
Divisors
Compac't Surfaces
of S
Let S be a compact Riemann surface. A divisor D on S is a
whose values arc integers and which is non-tero only at finitely many points
of S. Addition of two divisors D1 and D2 is defined by (D1 + D2)(p) =
D1(p) + D2(p). The degree of D, degD, is the sum of its values. We write
D1 D2 if D1(p) D2(p) for every point p eS.

Let q be a holomorphic differential of an arbitrary type on S. Fix a pOint
p eS and consider two local parameters
and z2 in a neighborhood of p,
both mapping p to zero. The mapping z1 , z2 is conformal at the origin and
has, therefore, a non-zero derivative at 0. We conclude from formula (1.2) in
1.4 that if the representation of in z1 has a zero of order n ( 0) at the origin,
then the representation of q in z2 also has a zero of order n at the Qrigin. Thus
the zeros of and their orders are well defined, being independent of local
parameters. Similarly, we infer that the poles of a mcrornorphic differential
and their orders can be defined in an invariant manner.
Let be a meromorphic differential on S with the zeros of order
at the
-
160
points
and the poles of order n1 at the points qj The
differential ip is the function which takes the value m1

qj, and vanishes dsewhere. Hence, deg El, = m, E '4.
El, of the
at
value
Suppose, in particular, that is a lunction on S. By Theorem 5.3, we then

and P2 are differentials of the same type, the quotient
have deg El, = 0. If
is a function. From the definition of the divisor of a differential we see
that deg D,11,,2 degD,1 deg D,2. It follows that the divisors of differentials
of the same type all have the same degree.
Given a divisor El, consider the family which consists of all meromorphic
functions f with El1 El, together with the function which is identically zero.
This family is a complex linear space. Its dimension is called the dimension of
the divisor El
by dim El.
If D = 0, i3e., D(p) = 0 for every p ES, the space consists of the constants,
and so
dim0= 1.
(5.2)
We also conclude that
dim El =0 if deg D >
(5.3)
0,
because the space then contains only the zero function.
5.5. RiemannRoch Theorem

We denote here by Q the complex vector space of all holomorphic quadratic
differentials on S. We fix a non-zero .,li eQ and write D, = D2. If q is an
is a meromorphic
arbitrary meromorphic quadratic differential, f =
function. From El, = D1 + El2 we see that q' is holomorphic if and only if
It follows that
D,
dimQ = dim(D2).
(5.4)
Exactly the same reasoning can be applied to (1,0)-differentials. If D1 is the

divisor of a holomorphic Abelian differential, we conclude that the space of
these differentials has the dimension dim(D1). On the other hand, by what
was said in 5.3, this dimension is equal to the genus p of S. Hence
dim(D1) =
(5.5)
p.
The results (5.4) and (5.5) can be put together with the help of a classical
result on compact surfaces (Springer [1], p. 264).
Theorem 5.4 (RiemannRoch Theorem). On a compact Riemann surface of
genus p, every divisor D satisfies the equation
-
dimD = dirn(D
D1)
degEl
+ 1.
(5.6)
161
of Quadratic Differentials
6.
Then, by (5.5) and (5.2),
Let us first apply (5.6) for

deg D1
+ 1, so that deg D1 = 2p
p=
1+
2. By our previous remark, we have
degD,,=2p2
for every meromorphic (1,0)-differential
Now let 42 be a meromorphic quadratic differential. Then

(1,0)-differential. From degD,, +
degD,, =
is a
degD,2 it thus follows that

4p
4.
(5.7)
In particular, every non-zero holomorphic quadratic

surface of genus p has 4p - 4 zeros.
The dimension of Q can now be readily determined.
on a
Theorem 5.5. On a compact Riemann surface of genus p, the space of holohas dimension 1 !f p = 1 and 3p 3 if p> 1.
morphic quadratic
PRooF. In the case p = I, the RiemannRoch theorem is not needed to
determine the dimension of Q. We saw in 4.1 that cover transformations are
+ nw2, m, neZ. Formula (3.5) shows, therefore,
translations z z +
that p is a holomorphic quadratic differential for the covering group G if and
= o(z) for allm and n. It follows that oeQ is

only if q(z + mw1 +
a bounded holomorphic function in the complex plane and hence a constant.
Conversely, every constant is a quadratic differential for G. We see that
dimQ = 1.
Next suppose that p> 1. We fix a holomorphic quadratic differential and
denote its divisor by D2. After this, we choose D = D2 in (5.6). Then, by
(5.4) and (5.7),
dimQ
dim(D2 D1) + 3p
3.
(5.8)
Now deg(D2
= degD2 degD1 = 2p 2 > 0. Hence the desired result
dim Q 3p 3 follows from (5.8) and (5.3).
If p =
0,
the space Q reduces to zero.
6. Trajectories of Quadratic Differentials

6.1. Natural Parameters
be a holomorphic quadratic lifferential on a Riemann surface S. We
assume that p is not identically zero, and regard as fixed throughout this
section. A point peS is said to be regular if (p(p) 0, and critical if o(p) = 0.
We showed in 5.4 tbat these are invariant definitions, independent of the
Let
IV. Riemann
162
representation of o. Critical points form a discrete set, and on a compact

Riemann surface there are only finitely many of them
p bea regular point and q h(q) = z a
0,
p mapping pto the origin. Since q(O)
the origin in which the two branches of z
single-valued. For a fixed branch of
every integral function
are
is then also single-valued( this neighborhood of the orgin and uniquely

dz
determined up to an additive constant. From the invariance of
under changes of parameter it follows that every is a function on S near p.
0 we conclude that there is a disc around the
From 1Y(O) =
origin which z ,
q ' w
maps injectively into the complex plane. It follows that

is a local parameter near p. From
dw2 = o(z)dz2
we see that with respect to w, the function representing the quadratic differential p is the function which is identically equal to 1.
We call w =
a natural parameter at p. An arbitrary natural parameter
at p is of the form w + constant. We see that in each case, near a regular
point the local representation of in terms of a natural parameter is the constant function
1.
There are natural parameters at a crItical point also. Suppose that p ES is

a zero of order n of (p. Again, let q h(q) = z be a local parameter near p
which maps p to the origin. Then there is a disc D(0, r) around the origil3 in
which p(z)
in
with *(z) 0. We fix a single-valued branch of
D(0,r). If n is odd, we cut D(O,r) along its positive radius I = {xIO x <
and fix a branch of z ,
in D(0, r) \ I; if n is even, no such cut is needed. In
either case
=
=L
is single-valued in D(0, r) \ I.
valued and 0 in a disc D(0, r1)
needed in the definition of to.)
In I)(0.r1), the function
z
+ c1z + "),
z , co(z) =
D(0, r). (Note
c0
2)12
0,
is
= ZO(Z)2
single-valued. Since it has the non-zero derivative

it is injective in a disc D(O,r2) D(0,r1). It follows that
is
at the origin,
q
is a
single-
that the cut I is no longer
local parameter near p. We now call
a natural parameter at p.
163
6. Trajectories of Quadratic Dilferentials
From
(4)1)2
we
obtain
(p(z)dz2 =
(ii
+ 2)2CndC2
(6.1)
In other words, near a critical point of order n, a holomorphic quadratic

has the representation * (1 +
in terms of the natural
parameter C.
The idea of associating natural parameters with a quadratic differential is

due to Teichmlfer [1J.
6.2. Straight Lines and Trajectories

A continuously differentiable mapping y of an open interval I into the
Riemann surface S with a non-zero derivative on I is called a regular path on
S. Near a point p e y(t) we introduce a local parameter q , h(q) z and write,
wIth a slight abuse of notation, z(t) =
(h o
y)(t). We assume that y does not
pass through any critical point of qi. The function t '

well defined on I, of course modulo 2ir.
If
arg(.p(z(t))z'(t)2) = 0 = constant
is then
(6.2)
at every point tel, we say that v is a straight line (in the geometry induced by
the quadratic differential (p). The condition for y to be a straight line is often
expressed in the form
arg(q(z)dz2) = constant
along y. It follows from the definition that a straight line does not pass
through a zero of (p.
We say that a straight line is horizontal if U = 0, and vertical if U = iv. The
straight line (6.2) is a horizontal line for the quadratic differential

(p. A
straight line (6.2) is called maximal if it is
properly contained in any
regular curve on which (6.2) is true. A horizontal trajectory is a maximal
Lorizontal straight line. Similarly, a vertical trajectory is a maximal 'vertical
straight line.
In natural parameters, the trajectories have simple representations. Let w
be a natural parameter near a point of From dw2 = p(z) dz2 and from (6.2)
it follws that locally, a horizontal straight line is a euclidean horizontal line
segment in the w-plane. Similarly, a vertical straight line is locally a euclidean
vertical segment.
Near a critical point pe S the behavior of trajectories is more complicated.
w = 4)(z), be a natural parameter in a neighborhood of p.
Let =
Horizontal lines near p are horizontal line segments in the w-piane, whereas
in the C-plane, they are located in n + 2 different sectoral domains. Consider,
in particular, a horizontal line segment in the w-plane which contains the
Iv.
164
First-order zero
Regular point
Figure 10. Horizontal trajectories under natural parameters.
=
we conclude that its preimage on
S consists of n + 2 rays emanating from the critical point p such that the
origin. In view of the relation
angle between two adjacent rays is equal to 2n/(n + 2) (Fig. 10).

A trajectory with an endpoint which is a zero of is called critical. The
number of critical horizontal trajectories is countable, and on a compact

Riemann surface there are only finitely many such trajectories.
From the definition it is
that given a regular point of S, there exists
exactly one horizontal trajectory passing through that point.
6.3. Orientation of Trajectories

A sufficiently small .subarc of a horizontal trajectory is mapped by a natural
parameter onto a segment of the real axis. The orientation of the real axis can
thus be transferred locally to horizontal trajectories.
The global situation is more complicated. Let S0 = S \ {zeros of p}. For
is defined, we
any two natural parameters w1 and w2 of S0 for which w2 o
have w2 o
(z) = z + constant. The trajectory structure of is said to be
orientable if has an atlas of natural parameters w1, such that every change
of variables is of the form
wr1(z)
More briefly, we then say that
is
= z constant.
orientable.
Let us assume that p is not orientable. We shall show that S then has a
two-sheeted covering surface branched over the zeros of of odd order,
such that the lift of to is orientable.
on the punctured
In order to prove this, we consider for a moment
i = 1,
surface S0. Suppose that q, is a collection of holomorphic functions
..., defined in simply onnected domains
S0 with local
Then
=
on
Consider all triples (p.;, cxi), where p E U1 and is holomorphic and satiswith
if p = q and cc1d; =
We identify
fies
= on
denote
the
set
of
identified
triples
and ir:
S0 the
Let
at p = q.
projection which maps (p. z, x,) to p.
2,
165
6. Trajectoriesof Quadratic Differentlais
is an
It follows that
We introduce the standard topology on
unlimited covenng surface of S0. It becomes a Riemann surface with the
conformal structure lifted from S0.
Let g be the natural extension of
to the holes over the points of S \ S0.
More precisely, if the lift of' a closed path on S0 around a zero p of q'
terminates at the initial point, we have two points of ovar p (as we have over
2 over p. The latter
all points of S0). Otherwise has a branch point of
alternative occLrs if and only if p isa zero of odd order. If

the natural extension of 7r:
-.
then
is
it) is a two-sheeted branched
covering surface of S.
Finally, let and denote the lifts ofq' and to From the construction
it is clear that the functions form a holomorphic Abelian differential a and
that = a2. As the square of an Abelian differential

wished to show.
is onentable, as we
6.4. Trajectories in the Large
In order to study the global structure of trajectories, we choose a regular

point Po ES, fix a branch of
in a neighborhood of Po' and normalize it so
= 0. Let r
be the largest number, such that the analytic
that
continuation 10 of the local inverse of
maps the' disc D0 =
D(O, r)
injectively
into S. The image V0 = f0(D0) is called a maximal disc around Po' and
r = r(p0) is said to be its radius. The maximal disc V0 is uniquely determined
by qi, i.e., V0 does not depend on the choice of the integral function of
The function p r(p) is continuous in p.
Next we choose a point u1 D0 which lies on the real axis R; then f0(u1) =
of D1 =
Pt is a regular point. Hence, there is a conformal mapping
D(u1, r(p1)) onto the maximal disc V1 around p1. such that f1 = 10 in D0
and that f1 is the inverse of b1. By continuing this procedure, we obtain

such that =
connected chains of discs D0, D1, ..., D with centers on
is a conformal mapping of D1 into S and that
continuation off1.
is a direct analytic
Let G be the union of all the discs of such chains which we obtain
by starting from D0. Since the intersection of two chains is connected and
contains D0, the analytic continuation f off0 to G is single-valued. We write
1
G R and deduce that f(1) is the horizontal trajectory which passes
through the point Po
If [a, b] is a closed subinterval of 1 on which f is injective, there is a

rectangle {u + ivla u b, v } which f maps injectively into S.
The image of every horizontal line segment in this rectangle is a subarc of
some horizontal trajectory.
We shall now show that the character of the horizontal trajectory f(1)
varies, depending on whether f is injective or not on the whole interval 1.
P1. Riemann Surfaces
166
6.5. Periodic Trajectories

bE K such that f(a) = f(b). Among the
Suppose first that there are
pairs of points of [a, b] at which I takes the same value, there is a pair u0, u1
with a minimal distance from each other. It is not difficult to show that I is
then periodic. in G, with the primitive period w = u0 (cf. Strebel [6],
p. 39). In this case f can be continued analytically to the whole real axis by
= f([u0, u1]) f([O, ]) is a
periodicity. The horizontal trajectory
closed curve. There is a maximal rectangle R0 = {u + ivIO u w,v1
V2 } in whose interior f is analytic.
v
If f is injective in R0, then
maps the annulus A

<
conformally onto a ring
domain on S. The image is called the maximal annulus around the horizontal
trajectory Every circle
constant in A maps onto a closed horizontal
trajectory of S. It follows that the maximal annulus around is swept out by
closed horizontal trajectories, freely homotopic to and all of the same
length co in the w-plane. From the maximality it is clear that if ct1 and c2 are
closed horizontal trajectories of S, their maximal annuli are either disjoint or
identical.
1ff is not injective in the rectangle R0, simple reasoning shows that I has
another primitive period co'
and that f has an analytic extension throughout the complex plane C (cf. Strebel [6], p. 41). The parallelogram with
vertices at the points 0, co, co + co' and co' and with the opposite sides identified is mapped by f bijectively onto S. We conclude that S is a torus. The pair
(C, I) is a universal covering surface of S. From the global representation
q(z) dz2 it follows that the straight lines on S are images under f of
dw2
euclidean straight lines in the plane. All horizontal trajectories are closed
curves on S.
6.6. Non.Periodic Trajectories

Let us now assume that the function f, obtained by analytic continuation of
Then
a germ of 'b', is injectivc on the interval I Gr- P
f: I -. S is a parametric representation of the horizontal trajectory passing
through the point eS with which we started. The trajectory is now an open
arc, and we define its length to be the same as the length of I. (The metric
will be studied in the next section 7.) The two parts intp which
We denote & = fffo, u,1,)),
trajectory rays from
Po divides
Let L be the limit set of the ray
i.e.,
L is the set of points p eS for which
167
6. Trajectories of Quadratic Differentials
there exists a sequence of points
limit set L is contained in the
I tending to
such that f(u1) + p. The
of a, and it does not depend on the
choice of the initial point Po of a.

Regarding the set L, there are three essentially different possibilities. First,
tends to the boundary of S, and call
L may be empty. We then say that
are
boundary rays, the trajectory is said
and a
a boundary ray. If both
to be a cross-cut of S. On a compact surface, there are no boundary rays.
Second, L might consist of a single point p. The point p cannot be regular,
Hence, f(u) tends towards
because f could then be continued on past
is called a critical
on I. In this case,
the critical point p along
as u
ray.
The remaining case is that L contains more than one point. Suppose p EL
is a regular point, and consider the horizontal trajectory cx,, through p. This
cannot be closed, because it could then be covered by an open annulus which
contains only closed trajectories. thoose another point q e and construct
an. open rectangle R0 such that f,,1R0 is injective and that the image of the
middle line of R0 contains, the trajectory arc from p to q. Since p E L, we have
e R0, the horizontal line segment in
points p1 E a, such that
p. If
R0 through f,,'(p1) maps on a subarc of a. There are infinitely many such
subarcs, and we conclude that a has infinite length.
The same reasoning shows that if the subarc of cc,, from p to q has length
a) e cc with properly chosen signs
a and if p1
p. then q1 =
converge to q. Here q e a,, was arbitrarily chosen. We conclude that if p E L,
then the whole trajectory through p belongs to L.
The trajectory a has infinite length also in the case when p L is a critical
point. The ray
cannot end at p, because L would then reduce to the single
point p. Therefore, at least one of the finitely many sectors into which the
horizontal trajectory rays emanating from p divide a neighborhood of p

contains infinitely many points p1 a. After this, the reasoning used in th
case where p was a regular point can be modified so as to yield the desired
result (Strebel [6], p. 44).
contains more than one point,
Consequently, if the limit set L of the ray
then
is always of infinite length. The ray
is then said to be divergent.
Suppose that the initial point Po of
belongs to the set L. From what we
just proved it follows that the whole trajectory a is then contained in L. Since
L lies in the closure of a, we conclude that in this case
L=i
L is called recurrent. A trajectory both rays
A trajectory ray
with
of which are recurrent is said to be a spiral, and its limit set L is called a spiral
set (Fig. 11).
On a compact Riemann surface, every divergent ray is recurrent and all

non-periodic
are spirals, save for finitely many exceptions. (For
two simple examples, see Figs. 11 and 12. For a detailed account of trajectories on compact surfaces, we refer to Strebel [6], 11.)
Iv. gjeinann Surfaces
168
Figure 11. A torus and a part of its universal cover ng surface. A straight line projects
on a spiral.
Figure 12. Horizontal trajectories on a surface of genus 2. 4p 4=4 critical points
7. Geodesics of Quadratic Differentials

7.1. Definition of the Induced Metric
As in the previous section, qi is a holomorphic quadratic differential on a
Riemann surface S, and not identically zero. The invariant differential
(7.1)
is called the line element of the metric induced by (p.

Let y be a curve on S locally rectifiable with respect to the eucidean metric
in any parametric plane. The length of y in the metric induced by can be
obtained with the aid of the following geometric reasoning. First, if y lies in a
maximal disc around a point p, the length of y is equal to the eucidean length
of the image of y under a natural parameter defined in a neighborhood of p.
An
y not passing through any critical points can be subdivided into
parts each one lying in a maximal disc. The length I(y) of y is the sum of the
lengths of these parts it is independent of the subdivision of y.
Since the differential (7.1) is an invariant on S, we can also define directly
1(y)
=J
169
7. Gwdesics of Quadratic Differentials
in terms of arbitrary local parameters. The length 1(y) is then well defined also
in the case where y passes through critical points.
The invariant tq,(z)I dx dy is called the area element of the (p-metric. Hence,
of 42.
the total area of S is the
In the following, is fixed, and notions like distance, length, and area,
refer to the (p-metric, unless otherwise stated.
The area of a compact surface S is always finite. If the genus of S is 1, i.e.,
ifS is a torus, its universal covering surface is the complex plane. The lifted q'
is then bounded in the plane and hence constant. It follows that the (p-metric
is euclidean, a result at which we arrived in a different manner in 6.5.
7.2. Locally Shortest Curves

The existence of a unique shortest curve joining two given points of S can be.
without difficulty if the points are close to each other. The curve
itself can be described geometrically by use of natural parameters.
Theorem 7.1. Every point of a Riemann surface has a neighborhood in which
any two points can be joined by a unique shortest curve.
PROOF. Let a point p eS be given and suppose first that p is regular. Let V
be the maximal disc around p and {wI IwI <r} its image under a natural
parameter w = (t)(z). Let V0 V be the preimage of
<r/2, and Pi' P2
arbitrary points of V0. Then the preimage Yo of the line segment connecting
'b(p1) and D(p2) is the unique shortest curve which joins p1 and P2 on S. For
let y
Yo) be an arbitrary curve on S which joins p1 and
If y s1ays in V,
then clearly I(y0) < 1(y). If y leaves V, then 1(y)> r> l(yo).
Suppose next that p is a zero of 42 of order n. We proved in 6.1 that if C is
a natural parameter near p, then

= (n +
42(C)
w = '1(e)
(7.2)
<r. Let V0 now be the preiinage of the disc
< 2_I
on S.
Then any two points Pt and p2 in V0 can be connected by a unique shortest
curve. This is either a straight line segment in the w-plane, or it is composed
of two radii in the a-plane which emanate from the origin. The former case
occurs if and only if Iarg arg C2 I <2n/(n + 2), where C1 and C2 are the
C-images of Pi and
These conclusions can be drawn from (7.2); for the
details we refer to Strebel [6], p. 35.
in a disc ICI
It follows from the above that if the shortest curve is the union of two
radii, both angles 9 between these rays satisfy the inequality
170
0>
(7.3)
n+2
This "angle condition" will be utilized in the study of globally shortest curves.
7.3. Geodesic Polygons

A curve y on S is called a geodesic if it is locally shortest, i.e., every point p
V, no arc
has a neighborhood Von S such that for any two points P1. P2 e
joining Pi and P2 in V is shorter than the subarc of y from p1 to P2 From the
results of 7.2, we know the local structure of a geodesic. It will turn out that
a geodesic is also globally the unique shortest curve in its homotopy class.
A geodesic polygon is a curve which consists of open intervals of straight
lines (in the geometry of and of their endpoints. The endpoints can be
zeros of g. In case the polygon is a Jordan curve, it is called simple and
closed.
Assuming the existence of a geodesic, we shall first prove that it is uniquely
extremal in its homotopy class. The proof uses the argument principle in
its generalized form, in which the holomorphic function considered in a
subdomain of the complex plane is alhwed to have zeros on the boundary of
that domain.
Argument Principle. Let f be holomorphic in the closure of a plane domain A
bounded by finitely many piecewise regular curves. Let denote the arcs into
which the zeros E 9A off divide the boundary, and the interior angle at
bet ween the arcs
and Yj. Then
J
where
rn
d arg
f(z)
d arg
f(z) =
in1 +
are the orders of the zeros off in A and nj the orders of those on OA.
If f(z) s& 0 on 8A, this is the standard principle of argument. The refinement says that the zeros on the boundary have the weights
Teichmller ([1], p. 162) drew the following conclusion from the Argument.
principle.
41
be holomorphic l,i the closure of a

Lemma 7.1 (Teichmiiller's Lemma).
:nded by a simple closed polygon in
domain A in the complex plane which
the
at the vertices. If and denote
the co-metric, whose sides
the orders of the zeros of o in A and on 8A, respectively, then
(flj +
2 + sm,.
(7.4)
171
7. Geodesics of Quadratic DifTeren.tials
PROOF. On y, we have arg(q,(z)
dz2) = constant, and so
d ai'g q,(z) + 2d(arg dz) = 0.
(7.5)
The argument of the tangent vector dz increases by 2ic

Oj) after a
This observation, coupled with (7.5) and the Argument
full turn along
principle, yields
which is (7.4).
It follows from (7.4) that
(1
2.
We conclude that there are at least three angles

6, <
(7.6)
so small that
+2
(7.7)
Hence, these angles do not satisfy the angle condition (7.3).
7.4. Minimum Property of Geodesics

In order to prove that a geodesic is globally the unique shortest curve in its
homotopy class, we need two auxiliary results. As before, we assume that we
have a fixed 4,-metric.
Lemma 7.2. Let S = G be a simply connected domain in the complex plane and
and 22 points of G. Then there exists at most one geodesic from to 22.
us assume that there are two geodesics joining z1 and z2 in1G. If
they do not coincide we can find two subarcs, both from a point a to a point
b, which form a simple closed polygon. The angle condition (7.3) is satisfied
at the vertices, except possibly at the two points a and b. This is in contradiction with the fact that (7.7) holds for at least three angles.
0
PROOF. Let
A geodesic is called maximal if it is not a proper subset of any other

geodesic.
Lemma 7.3. In a simply connected subdomain of the complex plane every

maximal geodesic is a cross-cut.
PRool. Let y be a maximal geodesic in a simply connected plane domain

S = G. Fix a point 20 E y and represent a ray of y with the initial poini 20 by
172
P/. Riemann Surfaces
using its arclength u as parameter, 0 u < u,. Assume that y(u) does not
such that
tend to t3G as u
Then there is a sequence of points ,
=
' z C. By Theorem 7.1, there is a disc U around z in which any
two points can be joined by a unique shortest curve. Consider the maximal
subarc Yk of y which contains the point
and lies in U. There is a point
zne U, n > k, which is not on Y& Otherwise y would terminate at z, which
contradicts the fact that every geodesic arc in U can be continued to U.
is a geodesic which leaves U. On the
Therefore, the part of y from 2k to
other hand, there is a shortest curve and hence a geodesic from ; to ; inside
U. This is in contradiction with Lemma 7.2.
0
With the aid of the above two lemmas, the unique extremality of geodesics
can now be established.
Theorem 7.2. Let S be a Riemann surface and p and q points of S. Then a

geodesic arc from p to q is strictly shorter than any other curve in its homotopy
class.
PROOF. Let Yi and 12 be homotopic curves on S joining p and q. Let z0 be a
point of the universal covering surface D of S over the point p. By Theorem

2.1 (Monodromy theorem), the lifts ofy1 and 12 from z0 terminate at the same
point z over q. Since lifting does not change lengths, 'we may assume that S is
a simply connected domain D in the complex plane.
Let y be a geodesic from z0 to z; by Lemma 7.2, y is unique. Consider an
arbitrary curve y' in D which connects z0 to z. Replaciig subarcs of y' by

locally shortest arcs with the same endpoints does not nake y' longer. We
may thus assume that y' is a_geodesic polygon. Also, it is not difficult to
construct a Jordan domain G, G c D, which is bounded by a geodesic polygon
and which contains y and y'.
Suppose first that y is a straight arc; we may assume that it is horizontal.
Let z1, z2, ...,
denote the points of y which lie on a critical vertical arc
with respect to G. We pick an arbitrary point z' of an open interval (z1...1 ,
of y, where i can take any value 1,2,..., n;; = z. By Lemma 7.3, the maximal
vertical arc through z' is a cross-cut of G. It follows that y' intersects P.
Now let z' run through all points of (z1_1, z1). The vertical arcs fi then
sweep out a simply connected domain A1, which is mapped by 1 onto a

vertical parallel strip. If a1 denotes the width of the strip, then the length of
A1 is at least a1, with equality if and only if y'
A1 is a single arc parallel
to y. The domains A. are disjoint so that l(y')
= 1(y). Equality can hold
only if l(y' A1) a for every i. Starting with i = 1, we first deduce that then
A1 = y A1, and continuing the reasoning we conclude that y' = y.
After this, let y be an arbitrary geodesic. Then y is the union of straight
arcs
For every
we construct the orthogonal strips
as before. An
arc orthogonal to does not meet y again; this follows from Lemma 7.2.
Neither can a intersect a orthogonal to Y&. k j. For if it would, we
Differentials
7.
173
would get a simple closed geodesic polygon with two interior angles equal
and the other
to it/2, one positive angle at the intersection of and
the condition (7.3). This would contradict the inequality
angles
are disjoint, and since y' goes through every
(7.6). Therefore, the strips
we conclude as before that 1(y') 1(y) and that equality can hold only if
it follows from Theorem 7.2 that every horizontal arc is uniquely length
minimizing in its homotopy class.
7.5. Existence of Geodesics
are arbitrary, it is not always

If the Riemanu surface S and the
possible to connect two given points p and q with a geodesic. A simple
example is the case in which S is a non-convex plane domain and the metric
is euclidean. The existence of geodesics can be shown if the distance between
the lifts of the points p and q to the universal covering surface is smaller than
their distances to the boundary.
In view of our applications, we shall restrict ourselves to the case in which
S is a compact surface. Then every point of the universal covering surface D
has an infinite distance to the boundary of D (Ahlfors [1]).
This is trivially true if D is the complex plane. For then 3D = {co}, and we
know that the q'-metric is euclidean (c1 6.5 and 7.1).
If D is the unit disc, the Dirichiet regions of the covering group of D over
S are relatively compact (Theorem 5.1). Hence, given a point eD, there is an
r0 < 1 such that the disc Iz I
covers the Dirichiet region with center at
Pick an r1 such that r0
<1, and let d denote the
between the
=
r1
can
be
covered
by
the images
Fhe
circle
IzI
=
circles IzI = r0 and
of <r0 upder finitely many cover transformations. The images of the disc
z <r1 under these finitely many transformations are contained in a disc
IzI <r2 with r1 <r2 < 1. From the invariance of q-distances under the covering group we conclude that the distance from to the circle Izi = r2 is 2d.
A repetition of the argument shows that the distance from to
is infinite.
We can now prove the existence of geodesics on compact surfaces.
I
Theorem 7.3. Let S be a compact Riemann surface and p and q points of S. Then
each homotopy class of curves joining p and q on S contains a unique shortest
(hence geodesic) arc.
PRooF. As in the proof of Theorem 7.2, we may replace S by its universal

covering surface D. Let two points z1 and z2 of D be given. Since the distance
from z1 and z2 to 3D is infinite, we can find a Jordan domain G, G c D, such
that
z2 G and that any arc connecting z1 and z2 in D and leaving G
cannot be length minimizing. If a denotes the infimum of the lengths of the
IV. Riemann Suhaces
174
curves in D which join z1 and z2, we then obtain the same infimum a if we
restrict attention only to curves which lie in G.
a.
Let (y1) be a minimal sequence of curves in 6 from z1 to 22, i.e.,
into n equal parts and take n so
Subdivide the parameter interval [0,
large that the endpoints of the resulting subarcs of y, can be joined by a

unique shortest arc in D. That this is possible follows from Theorem 7.1,
combined with a standard compactness argument. For a subsequence (y1j,
these n + I endpoints converge. By joining the limit points with shortest arcs
in D we obtain a shortest arc y from 21 to 22. Being globally shortest y is also
locally shortest, i.e., a geodesic.
The uniqueness follows from Theorem 7.2.
0
7.6. Deformation of Horizontal Arcs

A horizontal arc of S is shortest in its homotopy class. We shall prove that
this is asymptotically true if the. competing curves are images of under
deformations of S (Teichmller [1], p. 159).
Lemma 7.4. Let S be a compact Riemann surface, f: S ' S a homeomorphism
homoto pie to the identity, and a horizoNtal arc. Then there is a constant M,
which does not depend on such that
2M.
h: S x [0, 1] S be a homotopy from the identity mapping to f.

Fix a point peS and denote by the path t
Let be the (unique)
geodesic in the homotopy class of
If p' is close to p. the difference
is majorized by the sum of the distances between p and p' and
PROOF. Let
f(p) and f(p'). Hence, the function p

it follows that
is continuous. Since S is compact,
<
M=
peS
Now let p be the initial point and q the terminal point of the horizontal arc
denotes the path t Vq(l t), then
is homotopic to By
Theorem 7.2, the geodesic is shortest in its homotopy class. Therefore,
+ 2M,
as we wished to show.
On a spiral trajectory we can take as long as we please. If is a part of a

closed trajectory, we may allow to cover itself. i'hus we can always let
and have then
1.
CHAPTER V
Teichm!ler Spaces
Introduction to Chapter V
In this chapter we
the theory of Teichmller spaces of Riernann

s:irfaces by utilizing the results in all four preceding chapters.
In section 1 we define the notion o! a quasiconformal mapping between
Riemann surfaces and prove that the existence and uniqueness theorems
for Reltrami equations generalize from the plane to Riemann surfaces. The
complex dilatation turns out to beat 1, 1)-differential on a surface. The uniqueness theorem shows that every such
differential determines a
conformal structure for the surface.
The Teichmller space of a Riemann surface is defined in Section 2 as a set
olequivalence classes of quasiconformal mappings. A metric is introduced in
this space in the same manner as in the universal Teichmller space. The use
of the complex dilatation leads to a characterization of the Teichmller space
in terms of different conforrwi structures.
In sections 3, 4 and 5 we consider Riemann surfaces which have a halfplane as a universal covering surface. In section 3, the quasiconformal mappings used in the definition of a Teichmller space are lifted to mappings of
the half-plane onto itself. This gives a clear picture of an arbitrary Teich-
muller space as a subset of the universal space, and makes it possible to

generalize many previous results.
In 111.4 we mapped the universal Teichmuller space homeomorphically
onto an open set in the space of Schwarzian derivatives, in section 4 of this
chapter, we consider
restriction of this mapping to the Teichmller space
of a Riemann surface. The image is then contained in the subspace consisting
of the Schwarzians which are holomorphic quadratic differentials for the

covering group of the half-plane over the "mirror image" of the given surface.
176
V. TeichmllerSpaces
The unifying link with the earlier results concerning the universal space is
provided by a theorem which says that the image of an arbitrary Teichmller
space is the inrersection of the image of the universal Teichmller space with
the space of the quadratic differentials for the covering group. It follows that
the imag is an open set in the space of quadratic differentials. The distance
from a point of this image to its boundary can be estimated. Using Schwarzian
derivatives, we also obtain simple estimates relating the metric of a Teichml
ler space and the metric it inherits from the universal space, showing that
these two metrics are topologically equivalent.
The results of section 4 are used in section 5, where a complex analytic

structure is introduced into Teichmller spaces. Quasiconformally equivalent
Riemann surfaces turn out to have isometrically and biholomorphically

isomorphic Teichmller spaces.
While sections 25 deal with the general theory of Teichmller spaces, the
remaining sections 69 are primarily concerned with Teichmller spaces of
compact surfaces. Section 6 is devoted to the study of the Teichmller space
of a torus, which is shown to be isomorphic to the upper half-plane furnished
with the hyperbolic metric.
In section 7 we consider extremal quasiconformal mappings of Riemann
surfaces, which determine the distance in the Teichmller space, i.e., which
have the smallest maximal dilatation in their homotopy class. A necessary
condition for the extremal complex dilatation is derived in the general case. If
the surface is compact, we can conclude that the extremal is always a Teichwhere
muller mapping, i.e., its complex dilatation is of the form
o k < 1 and is a holomorphic quadratic differential of the surface.
In section 8 we prove Teichmuller's famous theorem that on compact

Riemann surfaces of genus >1, every Teichmller mapping is a unique cxtremal in its homotopy class. In section 9 we show how this result leads to
a mapping of the Teichmller space of a compact surface onto the open unit
ball in the space of holomorphic quadratic differentials. If the surface is of
genus p (>1), the Teichmller space is proved to be homeomorphic to the
euclidean space
Finally, the connection between the Teichmller metric and the complex analytic structure is discussed briefly, and some remarks
are made on the Teichmller spaces of
surfaces of finite type.

1.1. Complex Dilatation on Riemann Surfaces
A homeomorphism f between two Riemann surfaces S1 and S2
is called
K-quasiconformal if for any local parameters of an at!as on S1, I = 1, 2, the
mapping h2 ofo
is K-quasiconformal in the set where it is defined. The
mapping f is quasiconformal if it is K-quasiconformal for some finite K 1.
177
Suppose that the local parameters h1, k1 of S1 have overlapping domains

U1, V1, and that f(U1 V1) lies in the domains of the local parameters h2,
h = k2oh1, we then have in
k2 of S2. Using the notation g = h1
k1(t11 n V1),
ho(h2ofoh')og.
The mappings h and g are conformal and, therefore, do not change the maximal dilatation. It follows that K-quasiconformal mappings between Riemann
surfaces are well defined.
be a universal covering surface of S1. where D is the unit disc or
Let (0,
the complex plane. (Here and in what follows the trivial case in which D is the
extended plane is excluded.) For the quasiconformal mapping f:

S2,
consider all mappings w = h2
where we choose
to be a suitable
Then the complex dilatations of the mappings w define a
restriction of
function p on 0. Set k1 = o' oh1, where g is an arbitrary cover transformation of 0 over S1. Then h2 of a kj1 w og. From this and formula (4.4) in
14.2 it fllows that p satisfies the condition
p=(pog)
(1.1)
for every cover transformation g. Consequently, a quasiconformal mappingf

of a Riemann surface S1 determines a Beitrami differential for the covering
group G or, what is the same, a Beltrami differential on the surface S1 (cf.
IV. 1.4 and IV.3.6). This differential is called the complex dilatation off.
We can also arrive at the complex dilatation of a quasiconformal mapping
of a Riemann surface in a slightly different manner; namely, by lifting the
given quasiconformal mapping to.a mapping between the universal covering
surfaces. Let (D, it1) be a universal covering surface of
i = 1, 2, and 4 the
covering group of D over S,. Consider a lift w: 0 i 0 of the given quasiconformal mapping f: S1 + S2. Since the projections it1 and are analytic local
homeomorphisms, w is quasiconformal. Let p be the complex dilatation of w.
Because w o go
is conformal for every g e (cf. IV.3.4), the mappings w
and wog have the same complex dilatation. Hence, we again obtain (1.1).
Clearly this p is the complex dilatation of 1.
We assumed that S1 and S2 admit
same universal covering surface D.
But if S1 and 52 are quasiconformally equivalent, the same is true of their
universal covering surfaces. Therefore, it is not possible that the universal
covering surface of one of the surfaces is a disc and of the other the complex
plane.
The existence theorem for Beltrami equations (Theorem 1.4.4) can be
generalized to Riemann surfaces.
Theorem 1.1. Let p be a Beltrami dWerential on a Riemann surface S. Then

there is a 4uasiconformal mapping of S onto another Riemann surface with
complex dilatation p. The mapping is uniquely determined up to a conformal
mapping.
v: Teichmuiler Spaces
178
PROOF. We consider
p as a Beltrami differential for the covering group G of D

over S. By Theorem 1.4.4, there is a quasiconformal mapping f: D , D with
complex dilatation ji. Since (1.1) holds, f and fog. have the same complex
dilatation for every p G. Then fog of1 is conformal, and we conclude that
f induces an isomorphism of G onto the Fuchsian group C' {Jo g
ge
G}. If it and it' denote the canonical projectipns of D onto S and S' = DIG',
of'
then q,o it
= it' of
defines
a quasiconformal mapping q, of S onto S'. This
mapping has the complex dilatation P.

Let cli be another quasiconformal mapping of S with complex dilatatio p
and w: D . D its lift. Then w
D
D is conformal, and so its projection
04,1 is also conformal.
.
-
1.2.
Conformal Structures
Let p be a Beltrami differential on a Riemann surface S, and h an arbitrary

local parameter on S with domain V. From the existence theorem for Beltrami equations it follows that there is a complex-valued quasiconformal mapping w of h( V) with complex dilatation po h'. (For this conclusion we can
use the plane version, Theorem 1.4.4.) Then f = w h is a quasiconformal
mapping of V into the plane with complex dilatation p. 1ff1 and 12 are two.
such mappings with intersecting domains V1 and V2, then by the Uniqueness
theorem (Theorem 1.4.2), 12 ofr1
important conclusion:
is
conformal in f1(V1
V2). This allows an
A Beltran*i differential of S defines a conformal structure on S.
the structure induced by p,

If H is the original conformal structure and
is determined by all quasiconformal mappings of open subsets of
then
(S. H) into the plane whose complex dilatations aie restrictions of p. These
mappings are conformal with respect to the structure fIn.
We can relax the conditions on p slightly and still obtain conformal structures. In fact, the above reasoning works if p isa ( 1, 1)-differential of S a*id
< I in every corn pact subset of S.
p
1
II
1.3. Group Isomorphisms Induced by Quasiconformal

Mappings
'Let us now
The lifts
the
a lift
it
lifted
that the universal covering surface D of S is the unit disc.

of Riemann surfaces need not possess limits at
However, if the homeomorphism is quasiconformal, then
admits a homeomorphic extension to the boundary. This makes
rephrase Theorem IV.3.5 in terms of the boundary behavior of
179
L Quasiconformal Mappings of Riemann Surfaces
Before formulating the theorem, we make a remark on the transformation

of the limit sets of covering groups. Let f: D ' D be a lift of a quasiconformat
mapping of a Riemann surface with the covering group G onto a Riemann

surface with the covering group G'. If z is a fixed point of g G and f(z) =
then (fog of' = f(g(z)) = We conclude that f maps the fixed points
of G onto the fixed points of G'. Since the limit set is the closure of the set of
the fixed points (1V.4.5), it follows that f maps the limit set L of G onto the
limit set L' of G'.
Theorem 1.2. Let S and S' be Riemann surfaces with non-elementary covering
groups G and G', (pt: S + S', i = 0, 1, two quasiconforma! mappings, andf0 a lift
of
Then
induce the same group isomorphism between G and G'
and
of coi which agrees withf0 on the limit set of G.
and only if there is a
first that there is a lift

such thatf1 = .10 on the limit
set L of G. Because f0 and j1 map L onto the limit set L' of G' and because L
is invariant under G, we then have
PROOF. Suppose
=f1ogof11,
geG,
(1.2)
at every point of L'. Both sides are Mbius transformations. Since they are
equal on a set with at least three points, they agree everywhere.
In order to prove the necessity of the condition, we now assume that (1.2)
is true in D. Settingh =
of1, we rcwrite(l.2) in the form
goli
hog.
If z is a fixed point of some g, then g(h(z)) = h(z), I e., h(z) is also a fixed point
of q. If z is an attractive fixed point and c D, then for the nth iterate of g,
* z as n ,
On the other hand,
=
-+ h(z). Hence
Ii(z) = z for all fixed points of G. Since these fixed points comprise a dense
subset of L(see IV.4.5), it follows that f0(z) =f1(z) for all z in L.
Theorem 1.2. combined with Theorem IV.3.5, plays an important role in

the theory of Teichmiier spaces. The following special case deserves particular attention.
Theorem 1.3. Let S be a Riemann surface with a non-elementary covering

group. 1ff: S S is d c'oP!formal mapping homotopic to the identity, then f is
the identity mapping.
By Theorem IV.3.5, f and the identity mapping of S induce the same

group isomorphism of the covering group of D over S. By Theorem 1.2, f has
a lift which is the identity mapping of D. Hence, the projection f itself is the
identity mapping.
0
S
180
V. TeichmlJer Spaces
1.4.
Homotopy Modulo the Boundary
For covering groups of the first kind, the existence of homotopy between two
mappings 4)o and q1 is equivalent to the existence of lifts which agree on the
whole boundary of D. in the theory of Teichmiler spaces this is a very

satisfactory state of affairs. For analogous behavior to occur when covering
groups are of the second kind, which would make it possible to develop a
unified theory, we need a stronger form of homotopy.
Let us consider a Riemann surface S = DIG, where D is the unit disc and
G is of the second kind. We denote by B the non-void complement of the
limit set L of 0 with respect to the unit circle. Then
= DIG LI BIG is a
bordered Riemann surface (ci. IV.1.3 and IV.4.5).
if we use the quotient representation DIG for Riemann surfaces under
consideration, an amazingly strong result can be easily proved: A quasiconforma! mapping 4) of S = DIG onto S' = DIG' can always be extended to a homeomorphism of S* onto (S')*.
In order to prove this, we consider a lift f: D - D of qi. We continue f by
reflection to a quasiconformal mapping of the plane. The extended f then
induces the isomorphism g -. fog
between G and 0' in the whole plane.
We proved in 1.3 that f maps the set of discontinuity of G onto the set of
discontinuity Q' of 6'.

We assumed that G is of the second kind, in which case G' also is of the
second kind; Extend the canonical projections it: D DIG and it': D ' DIG'
to the domains fI and Cl'. Then
(p*on = it'of
defines a
quasiconformal mapping p of the double Cl/G of S onto the double

of S'. Its restriction to
(D B)/G is the desired extension of (p.
S i S', I =0, 1, be two quasiconformal mappings between the RieLct
mann surfaces S = DIG and S' = DIG'. We just proved that
and can be
extended to mappings of
onto (5')*. We say that 4)0 is homotopic to
CZ'/G'
=
on the border and there is a homotopy from
to qi1 which is constant on the border.
modulo the boundary if
Theorem 1.4.
quasiconformal mappings (p,: S
modulo the boundary if and only if they can be
'
5', 1 = 0, 1, are homotopic

to mappings of D which
agree on the boundary.
Assume first that Qo and 4)i are homotopic modulo the boundary. If
then the lift 11 of
homotopic to 10 through the lifted
homotopy agrees with f0 on the set B. The mappings f0 and 11 determine the
same group isomorphism (Theorem IV.3.5). From the proof of Theorem 1.2
it follows that J0 = f1 on L.
on the boundary of D, we construct a homotopy f,
Conversely, ii
fromf0 tof1 as in the proof of Theorem IV.3.5, and conclude again that it can
PROOF.
fo is a lift of
I. Quasiconformal Mappings of Riemanu Surfaces
181
be projected to produce a homotopy from
to
Since J keeps every
point of B fixed, the projected homotopy is constant on the border of S. 0
1.5. Quasiconformal Mappings in Homotopy Classes

Not every sense-preserving homeomorphism between two given Riemann
surfaces is homotopic to a quasiconformal mapping. A trivial counterexampie is the case where one of the surfaces is a disc and the other the complex
plane. These are homeomorphic Riemann surfaces but not quasiconformally
equivalent.
In the case of compact surfaces, the situation is different (Teichmller [2J).
Theorem 1.5. Let S and S' be compact, topologically equivaleflt Riemann surfaces. Then every hoinozopy class of sense-preserving homeomorphisms of S
onto S' contains a quasiconformal mapping.
PRooF. Let f: S
S'
be a sense-preserving homeomorphism. Since S is com-
pact, it has a finite covering by domains

such that Uk is
U2, ...,
conformally equivalent to the unit disc and 3Uk is an analytic curve. Set
k = 1, 2, ..., n, as
= .1 and define inductively a sequence of mappings
in S\Uk, while in
the mappingfk is the BeurlingAhifors
=
extension of the boundary values fk..IIUk. More precisely, we map Uk and
(Ut) conformally onto the upper half-plane H.. Since Uk and .Ik-1 (Uk) are
Jordan domains, these conformal transformations of Uk and fk_j(Uk) onto H
have homeomorphic extensions to the boundary (see 1.1.2). We normalize the
mappings so that the induced self-mapping w of H keeps
fixed. After that,
we form the BeurlingAhifors extension of
in 1.5.3. By transferring
this extension to S we obtain fkl Uk. The
Uk is a dilTeomorphism
and hence locally 9uasiconformal. Moreover, if
is quasiconformal at a
point z e aUk, then
V is quasiconformal for some neighborhood V of
z (cf. [LV], pp. 8485). Hence, fk is quasiconformal at z, because 311k is a
removable singularity (cf. Lemma 1.6.1). It follows that I,, is a quasiconformal
mapping o S, since S is compact.
The mapping (p, t) . tfk(p) + (1
(p) is a homotopy between
and fk It follows that 1,, is homotopic to f.
0
Theorem 1.5 is not true for arbitrary Riemann surfaces S and S', not even
in cases in which S and S' each admits a disc as its universal covering surface.
We shall prove later (Theorems 4.5 and 6.3) that if S
S' are arbitrary
Riemann surfaces which are quasiconformally equivalent, then every homotopy class of quasiconformal mappings of S onto S' contains a real analytic
quasiconformal mapping. This result can be regarded as a generalization of
Theorem 111.1.1.
V. TeichmUl!er Space
182
2.
Definitions of Teichmller Space
2.1. Riemann Space and Teichmller Space

We shall now generalize the notion of the universal Tt chmller space intro-
duced in III.! and define the Teichmller space for an arbitrary Riemann
surface.
Let us consider all quasiconformal mappings f of a Riemann surface S

onto other Riemann surfaces. If two such mappings f1 and 12 are declared to
be equivalent whenever the Riemann surfaces 11(s) and f2(S) are conformally
equivalent, the collection of equivalence classes forms the Riemann space
of S.
In the classical case where S is a compact Riemaun surface we could

equally well start with homeomorphic mappings of S: by Theorem 1.5, every
homotopy class of homeomorphisms contains quasiconforml mappings.

study of R5 is called Riemann's problem of moduli. In the case where S
is the upper half-plane, the equivalence relation is so weak that all mappings
f are eqtivalent, and so R5 reduces to a single point.
Teichmller [1] observed that even in the case of a compact surface, a
space simpler than R5 is obtained if we use a stronger equivalence relation. Let
f1 andf2 be quasiconformal mappings of a Riemann surface S. Suppose that
the universal covering surface of S is the extended plane or the complex plane
or a disc with a covering group of the first kind. Then and 12 are said to be
equivalent if 12 ofr' is homotopic to a conformal mapping of f1(S) onto

f2(S). If the universal covering surface of S is a disc and the covering group is
of the second kind, i.e., if S is bordered, "homotopic" in this definition of

equivalence is to be replaced by "homotopic modulo the boundary".
Defirntion. The Teichmller space of the Riemann surface S is the set of the
equivalence classes of quasiconformal mappings of S.
Teichmflller restricted his interest to compact Riemann surfaces and, a

little more generally, to certain cases in which T5 is finite-dimensional (cf. 9.7).
Ahlfors [1] seems to have been the first to use the name "Teichmller space",
this in 1953. The above definition applying to all Riemann surfaces is due to
Bers ([7], [8]).
It is not difficult to see that if S = H, then T5 agrees with the universal

Teichmller space TH. In applying the above
T5 to S = H, we
first note that all. quasiconformal images of H are conformally equivalent. It
follows that we may consider without loss of generality only the normalized
quasiconformal self-mappings of H. which we denoted in III.! by f's. By Thebe homotopic.modulo the boundary
orem 1.4, the condition that fMz
to a conformal mapping is fulfilled if and only iffu2o(fhulVt agrees with the
identity mapping on the real axis R. Consequently, fe" is equivalent to
2. Definitions of Teichmfler Space
183
=
By the definition in
by the above definition if and only if
and fM2 to determine the same point in the
1111.1, this is the condition
universal Teichmller space TN.
A connection is obtained between
and the universal Teichmller spac.
if we lift the mappings between S and other Riemann surfaces to
between the universal covering surfaces. In this way we are able to transfer
many results associated with the universal Teichmller space to the general
case. This will be done in sections 3-5.
By slightly changing the definition of T5 we arrive at the reduced Teichmller
space of S. Its points are also equivalence classes of quasiconformal mappings
of S, but now two such mappings f1 and 12 are declared equivalent 1ff2 ofr'
is
just homotopic (not necessarily homotopic modulo the boundary) to a
conformal mapping.
The reduced Teichmller space differs from the Teichmller space T3 only
If S is bordered. In what follows, we shall not deal with reduced Teichmller
spaces. For this reason we henceforth use the term "homotopy" to mean
"homotopy modulo the boundary" in the case of bordered surfaces. This
simplifies the language and, if this convention is kept in mind, should not
cause confusion.
2.2. Teichmller Metric

Exactly as in the case of the universal Teichmller space, we define the
distance
r(p,q) =
(2.1)
between the points p and q of the Teichmller space T5 (cf. 111.2.1). In the
proof that r defines a metric in
the only non-trivial step is again to show
that r(p, q) = 0 implies p q. This can be deduced from the following resulL
Theorem 21. Letf0: S * S' be a quasiconformal mapping and F the class of all
quasiconformal mappings of S Onto S' homotopic to 10. Then F contains an
extremal mapping, i.e., one with smallest maximal dilatation.
PROOF. Let D be a universal covering surface of S. The theorem is trivial if V
is the extended plane or if D is the complex plane and S is non-compact. In
the case where D is the complex plane and S is compact, the theorem will be
proved in 6.4. Hence, we may assume that D = H is the upper half-plane (cf.
IV.4.l).
By Theorem 1.4, we can lift each fe F to a self-mapping w1 of H such that

all mappings w1 Igree on the real axis. The class W = {w1jf F} contains its
quasiconformal Inits. Hence, there exists a mapping WE W with smallest
maximal dilatation (cf. 1.5.7). The projection of w is the extremal sought in F.
V. Teichmller Spaces
184
q, and let F now

Given the points p, q E T5, we fix the mappings e p,
be the class of all quasiconformal mappings of f0(S) onto g0(S) homotopic to
the universal TeichAgain mimicking what was done in
g0
muller space (cf. 111.2.1) we conclude that
r(p,q) =
(2.2)
In other words, in determining the distance between Two points of T5 we can
always take the inlimum of maximal dilatations in a hom*opy class of

quasiconformal mappings between two fixed Riemann surfaces.
Theorem 2.1 says that the inf in (2.2) can be replaced by mm. Consequently,
if r(p, q) = 0, the class F contains a conformal mapping; and so p = q. After
this, it is
that (Ta, r) is a metric space.
The point which is defined by the identity mapping of S is called the origin
of T5. The origin contains all conformal mappings of S.
2.3. Teichmller Space and Beltrami Differentials

The definition of the Teichmller space T5 can also be formulated in terms of
the Beltrami differentials on S. Every quasiconformal mapping of S determines a Beltrami differential on S, namely, its complex dilatation. Conversely,
if p is a Beltrami differential of S, then by Theorem 1.1 there is a quasiconformal mapping of S whose complex dilatation is p. and by the uniqueness
part of Theorem 1.1, all such mappitigs determine the same point of 1's. Two
Beltrami differentials are said to be equivalent if the corresponding quasiconformal mappings are equivalent. Hence, a point of can be thought of as a
set of equivalent Beltrami differentials. The Teichmller distance (2.1) can be
expressed in terms of Beltrami differentials:
11(P
r(p,q) =
(2.3)
Let S admit the half-plane as its universal covering surface. Then geodesics
same description as in the universal TeichmUller space (see
in T5 allow
dilatation for the point p E T5,
Theorem 111.2.2): If p is an extremal
then
(1 + I IY(' I
(1
is ext remal for the point p1 =

t(p,,O) = tr(p,0).
IY
LIYIPI'
The arc t *
0 t 1,
(2.4)
p and
By using Theorem 3.1, to be established iu subsection 3.1, we can merely

repeat the proof of Theorem 111.2.2. We only have to make the additional
verification that p, represents a point of
Since p represents a point of T5,
185
2. Definitions of TeichmUfler Space
it is a Beltrami differential for the covering group G of the universal covering

surface over S, i.e., (p o g)g'/g' = p for every g G. From (2.4) we see immedithis condition. Hence [pj is a point of
ately that p1 also
The following generalization of Theorem 111.2.1 is immediate.
Theorem 2.2. The Teichmller space
is pathwise connected.
PROOF. The geodesic t -+ [p3 is a path joining the origin to the point p in T5;
the path r , [tt) of T5 also has this property.
0
-
2.4. Teichmller Space and ConformaiStructures

Let S be a Riemann surface with the conformal structure H and {h} an atlas
of local parameters belongingto H. 1ff is a homeomorphism of S onto itself,
then {h of -1 } is an atlas which determines another conformal structure of S.
We denote this structure by
and note that
does not depend on
the particular choice of the atlas on H,
It follows from the definition that f: (S, H) '
is a conformal map..
ping. Conversely, if H and H' are conformal structures of S and f: (5, H)
(S, H') is conformal, then H' =
We say that two conformal structures H and H' of S are deformation equivalent If can be deformed conformally to H', i.e., if there is a onformal mapping of (5, H) onto (S, H') which is homotopic to the identity.
In 1.2 we showed that every Beltrami differential p on the Riemann surface
(5, H) defines a new conformal structure
Given two conformal struc-
tures H and H' of S. suppose that there exists a quasiconformal mapping

(5, H'). Let p denote the complex dilatation of f. Then
f: (S, H)
H' =
(2.5)
This follows directly from the definitions, because now f: (S,

-+ (S, H') is
conformal.
There is a simple connection between different Structures
and points of
the Teichmller space of S.
Theorem 2.3. The conformal structures
H1
and H2 induced by the Beltrami
and #2 on the Riemaun surface S are deforma'ion equivalent

dWerentials
and only (I ILk and #2 determine the same point in the Teichmller space T5.
PROOF.
Let
I=
1,
2, be quasiconformal mappings of S with complex
dilatations
If q:
is a conformal mapping homotopic to
the identity, we first conclude that the mapping
v. 'reichmuhier Spaces
186
is conformal. AlsO, we see that 12

is homotopic to h. It follows that
are equivalent.
and
Conversely, if Pi and P2 are equivalent, there is a conformal map h:f1 (5) '
S is homotopic to the identity. In addi12(S) such that p = fj' oh of1: S
tion, (S, H1) -. (S. H2) is conformal, and so H1 is equivalent to H2.
0
We conclude that the Teichmller space can be characterized as the set of
equivalence classes of conformal structures H,, on S modulo deformation. Note
that a conformal structure H' on S is of the form H,, if and only if id: (S, H) -.
(S, H') is quasiconformal.
2.5. Conformal Structures on a Compact Surface

In considering different conformal structures on a swface S, we assume here
that for any two structures H and H', the identity mapping of (S, H) onto
(S; H') is sense-preserving.
a compact surface.
above results can then be supplemented ifS is
Theorem 2.4. On a compact Riemann surface S, every conformal structure is

deformation equivalent to a structure induced by a Beltrami
of S.
Let H be the given and H' an arbitrary conformal structure on S. By

Theorem 1.5, there is a quasiconformal mapping f: (S, H) -+ (S, H') which is
homotopic to the identity. Let f have the complex dilatation Then H' =
fjH_) (formula (2.5)). But f: (S, H,,) ' (S,f,,(H,,)) is a conformal mapping
homotopic to the identity. Consequently, H' =
is deformation equivalent to H,,.
0
PRoOF.
Theorems 2.3 and 2.4 yield an important characterization of T.

Theorem 2.5. The Teichmller space of a compact Riemann surface is isonwrphic to the set equivalence classes of conformal structures modulo deformation.
This result can also be expressed in somwhat different terms. Let
denote the set of all conformal structures of S. The group

consisting of all sense-preserving homeomorphic self-mappings of S acts on
If
HE .r(S) and fe
(S), then (H) e
Let Homeo0(S) be the subgroup of Homeo'(S) whose mappings are homo-
topic to the identity. Then H, H' e .*'(S) are deformation equivalent if and
only jf there is an fe Homeo0(S) such that
= H'. It follows, therefore,
that for a compact surface S we have the isomorphism
.it(S)/Homeo0(S).
187
2. Definitions of Teichm Her Space
In section 8 we shall prove that every class of equivalent complex dilatawhere 0 k < I and
tions contains a unique dilatation of the form
is a holomorphic quadratic differential of S. It thus follows from Theorems
2.4 and 2.5 that every conformal structure of a compact Riemann surface is
deformation equivalent to a structure induced by a Beltrami differential

Since q is uniquely determined up to a multiplicative positive constant (except of course for the case k = 0), we see that there is a simple
relationship between the normalized quadratic differentials and the equivalence classes of conformal structures on S.
The role of holomorphic quadratic differentials in the Teichmller theory
of compact Riemann surfaces will be studied in more detail in sections 79.
2.6. Isomorphisms of Teicbmiiller Spaces

In 111.5.2 we proved that the universal Teichmller spaces associated with
different quasidiscs are all isomorphic. Again, it is a trivial consequence of the
definition that we have a counferpart of this result in the general case.
Theorem 2.6. The Teichmller spaces of two quasiconformally equivalent

Riemann surfaces are isometrically bijective.
PRooF. Let S and 5' be Riemann surfaces and h a quasiconformal mapping
of S onto S'. The mapping f
is a bijection of the family of all
quasiconformal mappings f of S onto the family of all quasiconformal mappings of S'. If = f, o h', we have w2 o w1 = f2 of1'. We first conclude that
f1 and f2 determine the same point of if and only if w1 and w2 determine
the same point in Ti', i.e.,
{f)
{foh1]
(2.6)
is a bijective mapping of T5 onto T5.. It also follows that (2.6) is an isometry,

i.e., it leaves all Teichmuller distances invariant.
0
Under (2.6) the point {h] of is mapped to the origin of T5. We shall later
utilize this simple method of moving an arbitrary point of one Teichmller
space to the origin of another isometric Teichmller space.
If S and S' are compact Riemann surfaces of the same genus, they are
homeomorphic (IV.5.2). By Theorem 1.5, they are also quasiconformally
equivalent. We conclude from Theorem 2.6 that all Teichinller spaces of
compact surfaces of the same genus are isomorphic.
In sections 5 and 6 we shall introduce complex analytic structure in Teich-
muller spaces. We can then enhance Theorem 2.6 and prove that the Teichmuller spaces of quasiconforrnally equivalent Riemann surfaces are even
biholomorphically isomorphic.
V. Tcichmllcr Spaces
188
2.7.
Modular Group
Let h be a quasiconformal self-mapping of S. Then (2.6) defines bijective

isometry of onto itself. The group Mod(S) of all such isomorphisms [f]
[Jo h'] of T5 is called the modular group of Th.
IfS = H, in which case T3 is the universal Teichmllcr space, Mod(S) is the
universal modular group introduced in 111.1.2.
The modular group Mod(S) is also a generalization of the classical modular
group F of Mbius transformations acting on the upper half-plane If, in the
following sense: IfS is a torus, the Teichmller space can be identified with
H and the group Mod(S) with r'. This will be explained in 6.7.
In section 5 we shall prove that even in the general case, the elements of the
modular group are biholomorphic self-mappings of T5.
Let Qc(S) be the group of all quasiconformal self-mappings of S and Qc0(S)
the normal subgroup of Qc(S) whose mappings are homotopic to the identity.
We associate with every h Qc(S) the element [f] + a h1 3 of Mod(S).
This rule defines a mapping of the quotient group Qc(S)/Qc0(S) into Mod(S).
=
oh1eQc0(S), then
In fact, if
Clearly, this mapping
of Qc(S)/Qc0(S) into Mod(S) is surjective and a group homomorphism. We
remark that the mapping is injective if S admits no conformal self-mappings
other than the identity transformation. It follows that in this case the modular
group Mod(S) is isomorphic with the quotient group Qc(S)/Qc0(S). This is
also true of all Riemann surfaces quasiconformally equivalent to such an S.
The following result illustrates the homotopy condition which makes R5 a
quotient space of
jf
Theorem 2.7. The Riemann space is the quotient of the Teichmller space by the
modular group.
PROOF. Assume first that the points ff3 and fgj of are equivalent under
Mod(S). We then have a quasiconformal mapping h: S S such that Jo h'
is equivalent to g. But this means that there is a conformal mapping of f(S)
onto q(S), i.e., f
g determine the same point of R5.
Conversely, let f and g represent the same point of R5. Then a conformal
o of is a quasiconforinal selfmapping q,:f(S) , g(S) exists, and h =
qo(foh1)
mapping of S. From g =
we see that g and foh1 determine the
same point of
D
Theorem 2.7 says that two points [f1] and [12] of the Teichmller space
T5 are equivalent under Mod(S) if and only if the Riemann surfaces f1 (5) and
f2(S) are conformally equivalent. In other words, the modular group is transitive if and only if the Riemann space R5 reduces to a singleton. This occurs
only in the exceptional cases where quasiconformal equivalence of Riemann
surfaces implies their conformal equivalence. The universal Teichmuller space
(S = H) is such an exception.
3. Tcichm(iller Space and Lifted Mappings
189
Not only the Riemann space but even the Teichmller space may reduce
to a single point. This occurs if S is a sphere (a compact surface of genus zero)
or a sphere from which 1, 2 or 3 points are removed. All quasiconformal

images of S are then conformally equivalent; this can be seen if Theorem 1.4.4
is combined with the fact that three points of the extended plane can be
moved to arbitrary positions by a Mbius transformation. After this we
conclude that every point of T5 contains a conformal mapping, i.e., that T5 is
a singleton. If S is the sphere minus three points, S has the disc as a universal
covering surface (cf. IV.4.1), and the covering group is of the first kind.

3.1. Equivalent Beltrami Differentials
For a Riemann surface S, we defined the Teichmfler space T5 by means of
quasiconlormal mappings of S onto Riemann surfaces. Lifting these mappings
to mappings between the universal covering surfaces leads to new characterizations of T5 and makes it possible to see better the connection between the
general space and the universal Teichmller space.
We impose on the Riemann surface S the sole restriction that it has a

half-plane as its universal covering surface. Since we try to follow as closely
as possible the reasoning
in 111.1 and 111.2 in the case of the universal
Teichmller space, we take here the lower half-plane H' as the universal
covering surface of S. The cases in which the universal covering surface of S
is the complex plane will be discussed in section 6.
Given a Riemann surface S, we consider a Beltrami differential p on S or,
what is the same, a function p defined in H' which is a Beltrami differential
for the covering group of H' over S. As before, we denote by I " the uniquely
determined quasiconformal self-mapping of H' which has the complex dilatation p and which keeps fixed the points 0, I and
on the real axis R, and by
.i the quasiconformal mapping of the plane which has the complex dilatation
p in H', is conformal in the upper half-plane H and fixes the points 0, 1 and
Theorem 111.1.2 has an exact counterpart:
Theorem 3.1. The
fiR
Belt rami d ifferenrials p
or if and only
and v of S are equivalent ([and only
Let us first assume that p and v are equivalent. Let q, and be

quasiconformal mappings of S which lift to fP and ft', respectively. Then
there is a conformal map q: p(S) -+
such that oq is homotopic to i/i. By
Theorem 1.4, we have f" = h of" on the real axis R, where I:, as a lift of is
a Mbius transformation. Since
and [ both lix 0, 1, co, it follows that h
PROOF.
is the identity.
V. Teichmfler Spaces
190
f'
f'
conversely, that
the same isomorphism of the covering group of H' over S onto a
Suppose,
Fuchsian group G'. The projections of fM and f' map S onto the same
Riemann surface H'/G', and by Theorem 1.4, these projections are homotoplc. It follows that and v are equivalent.
=
by
if and only if
After this we can show that
=
repeating the proof of Theorem 111.1.2 verbatim.
0
3.2. Teichmller Space as a Subset of the Universal Space

Theorem 3.1 says that
and
are well defined injective mappings of the Teichmller space. In particular, T5
can be characterized as the set of equivalence classes
two mappings
being equivalent if they agree on k. We have thus arrived at the situation

which was our starting point in 111.1 when we defined the universal Teichmuller space. In the general case the complex dilatations of the mappings
is
are Beltrami differentials for the covering group G. If G is triyial, then
the universal Teichmlier space T (cf. also the remarks in 2.1).
This characterization of
that the family of Teichmller spaces
admits a partial ordering Let S1 and be Riemann surfaces and G1 and G2
the covering groups of H' over S1 and S2. If
is a subgroup of G2, then
In particular, every Teichmller space T5 can be regarded as a subset
of the universal Teichmller space T
Let r and r5 denote the Teichmller metrcs in the spaces Tand 1's. Then
the restriction nT5 is also a metric in 7's. From the definitions of r and r5 it
follows immediately that
(3.1)
It was for many years an open question whether the metrics VS and ri
actually agree. We now know that T5
not inherit its metric from the
and r T5 need not be the same.
This was proved by Strebel [4] who gave two examples of surfaces S for
In one case S is a punctured torus, in the
which r is strictly less than
othcr a compact surface of genus 2; cf. also 3.7 and 7.6.
Even though (3.1)does not always hold as an equality, the metrics tJ T5 and
t5) (T, r)
are topologically equivalent. In other words, the inclusion
is a homeomorphism onto its image. This will be proved in 4.6..
univerEal Teichmller space: The metrics
3.3.
of Teichmller Spaces
Lemma L22 is true in
Teichmllcr space T5:

a subsequence whose points
=
exists almost everywhere,
have representatives p, such that Jim
A Cauchy sequence in (T5,
191
Space and Lifted Mappings
3.
[f,j in the rrmetric,
f,,(z) uniformly itt the spherical metric,
and
' f"(z) locally uniformly in H' in the euclidean metric.
The proof is word for word the same as in Lemma 111.2.2. In this case every
p,, is a Beltrami differential for G, i.e., (p,t o g)g'/g' = p.,. From p,t(Z) -+ p(z)
differential for
ahnost everywhere it follows that the limit p also is a
G, i.e., [p3 is a point of
From this observation we obtain a generalization of iheorem 111.2.3.
Theorem 3.2. The Teichmller space (T5, t5) is complete.
For an application in section 9 we need the following result.
Lemma 3.1. Let [p,j
[v]in
k < 1, and p,t ,
[p3 in
a.e. Then [p3 =
T5.
PROOF. Let .%,tE [p,t]
be an extremal complex dilatation for which r5([p,t3,
(formula (2.3)).' The hypothesis [p,t]

[p3) artanhfl(2,t p)/(l
By Theorem 1.4.6, f,,,,, f,, and
[p3 then implies that A,, . p in
-+
= JJH it follows that H = f,IH, and so [p3 [vJ.
Since
3.4. Quasi-Fuchsian
Groups
The mappings lead to discontinuoufr groups of

with an invariant domain different from a disc.
Theorem 3.3. The mapping g
ing group
onto a group
defInes an isomorphism of the cover./, o g

of Mobius transfornuitions acting on the quasidisc
PROOF. Consider the quaaiconformal mapping
G, of the plane.
It is conformal in f,(H)1 because

is conformal. Since p is a Beltrami,
differential for G, the mappings and o g have the same complex
is conformal in
also. The common
tion. It follows that o g
boundary of f(H) and fM(H'), being the image of the real axis under fM' is a
quasicircie. We conclude, therefore, from Lemma 1.6.1 that o go
is a
Mbius transformation.
0
By the terminology we adopted in IV.4.6, the group
GM
quasi-Fuchsian. A quasi-Fuchsian group of this special type is called a

quasiconfonnal deforipation of the Fuchsian group G. Such groups were
is
discovered by Bers (4].
is a half-plane if and only if p is a trivial

The invariant domain
complex dilatation, i.e., p is equivalent to the complex dilatation whih is
192
Figure 13. Invariant
domain which is not a quasidisc.
identically zero. If G is of the first kind, the limit set of G,, is the whole
By Theorem 1V.4.2, only two strikingly different cases are
boundary of
then possible: Either the boundary is a straight line or else it is a quasicircle
which fails to have a tangent at each point of a set dense in the curve.
Not all simply connected invariant domains of properly discontinuous
groups of Mbius transformations are necessarily quasidiscs. A simple example is obtained if we consider a countable set of circles, all of diameter 1,
of which one has the center at 0 and the others at the points I + ,i, n 0, 1,
2
The method of Klein. which we described at the end of JV.4.6, applied
to this family of circles yields an invariant Jordan domain whose boundary
has a cusp at the origin. Thus the boundary curve violates the condition
and z3 -+ 0) and cannot be a
(6.1 1) of Theorem 1.6.7 at the origin (z2 = 0,
quasicircie (Fig. 13).
There can even be invariant domains whose boundary has positive area
(Abikoff [1]). In 4.3 we shall exhibit a general (albeit implicit) method
producing invariant domains which are not quasidiscs.
A point [it]
uniquely determines the domains A,, = f,,(H) and
can be regarded as a collecLike the universal Teichmller space,
tion of the quasidiscs AM (ci. 111.1.5). In 111.4 we defined a distance between
two such domains by using Schwarzian derivatives. The relation of this

distance to the Teichmller distance will be studied in section 4.
3.5. Quasiconformal Reflections Compatible with a Group

We show
determines uniquely the quasicircie
here that fM(R) admits always quasiconformal reflections which are compatible with the
group G.
The point [js] = p
Li order to make this statement more precise, we choose a /LEP and

consider the mapping
193
(3.2)
as before j denotes the reflection z
This mapping is a quasiconformal
reflection in
By Theorem 3.3, the mapping j induces an isomorphism of the group 6

acting on H' onto the quasi-Fuchsian group G,, =
acting
on
But since the elements of G and
are restrictions of Mbius transformations, we may consider g and =
in H and A, also. If
then
=
=
We conclude that the sense-reversing quasiconformal mappings 2 and

have the same complex dilatation. Therefore, if K, =
denotes this
complex dilatation, then
(K)
(3.3)
for every
e G,1. in other worth, the complex conjugate of the co,nplex dilatation of 2 is a Bhrami differential for the quasi-F uchsian group
Conversely, we prove that if 2 is a quasiconformal reflection in
whose
complex dilatation satisfies the relation (3.3), then 2 is the form (3.2) (cf.
Lemma 1.6.2). Set I
in the closure of H, and I =
in H'. Then f is
a quasiconformal mapping in the plane which is conformal in H. For
G,
we have in
of
Since fog of 1 =
formal self-mapping of
in
that J og of ' = everywhere. We see that f is a mapping
f,4, and so A. = fojof1 is of the form (3.2).
is
a con-
it follows
equivalent to
The point [u] e determines the conformal mapping f,41H uniquely, but
not
and hence not 2. In 4.8 we shall
that for each G and [p1, there
are Lipschitz-continuous reflections (3.2). We also remark that by Theorem
4.5 (to be proved in section 4), there are reflections A which are real-analytic
in i.hc complement of
3.6. Quasisymmetric Functions Compatible with a Group

The mapping fP induces an isomorphism of the group 6 onto the Fuchsian
group
G't=
In particular,
Mbius transformauon.
agrees with the restriction to R of a
Let us consider again the space X of normalized quasisymmetric functions,
194
V. Teichmufler Spaces
which we defined in 111.1.1 and studied in 111.3. We relate X to the group G
as follows: X(G) is the subset of X for whose functions h the composition.

hog a h' is the restriction to of a Mbius transformation for every g G.
We let X(G) Mherit the metric of X,
distance in X(G) is the restriction
to X(G) of the distance function p on X. Then (X(G),p) is a metric space, and
if G is trivial we get back our previous space (X, p).
Theorem 111.3.1 states that
the universal Teichmller space onto X. In order to generalize this theorem to and X(G), we need the following result.
Theorem 3.4. Every quasisymmetric functibn he X(G) has a quasiconformal

extension to the half-plane, such that the mapping fog of is conformal for
every geG.
The confonnal mapping fog of -' is of course the restriction to the halfplane of the Mbiustransformation which agrees with hog o
on the real
axis.
Theorem 3.4 is a deep result which has been established in steps. First, let
G be the covering group of a compact surface. Then Theorem 3.4 follows if
the classical result that he X(G) can be extended to a homeomorphic selfmapping of the half-plane compatible with G is combined with Theorem 1.5.
Kra [1] generalized this result by proving Theorem 3.4 for all finitely gener-
ated groups 6. The author showed that Theorem 3.4 holds for all groups
provided the quasisymmetry constant of h does not exceed
In full
Theorem 3.4 is due to Tukia [1]. Tukia's proof is so long
that to includcit here would unbalance our presentation. For this reason, we
content ourselves with a reference to Tukia's paper.
Quite recently, the
a forthcoming paper by Douady
and Earle [1] which
another proof of Theorem 3.4. This proof is
more explicit than thitof'I'Ukia. It is possible'to verify that, like the Beurling
Ahlfors extension, the DcMadyEarle extension possesses the following three
properties: 10 It is a diffboniorphism. 2 It is Lipschitz-continuous in the
hyperbolic metdcf
3 Its maximal dilatation has a bound
which depends only o&tb.c quasisymmetry constant of the boundary function.

Thanks to these thttc properties, the DouadyEarle extension can be used
in applications Instead of the BeurlingAhifors extension. The additional

property of the
extension of being compatible with the group
action leads to impOrtal3t new results. The proof of the contractibility of the
universal Teichrnller
in 11L3.2 was based on the use of the Beurling
Ahifors extension. Application of the DouadyEarle extension in its place
makes it possible to solve a long outstanding problem (DouadyEarle til):
Every
is
contractible.
In section 4 we shall see that Theorem 3.4, no matter how it is proved,

reveals remarkable properties of Teichmller space. Whenever possible, how-
195
ever, we shall establish such results by other methods, so as to render the

proofs self-contained.
Using Theorem 3.4 we can now generalize Theorem 111.3.1.
Tbeorem 3.5. The mapping
(3.4)
is a homeomorphism of (T5, t5) OfltO (X(G), P).
is well defined and injective. By

PROOF. By Theorem 3.1, the mapping
Theorem 3.4, it is surjective.
By Theorem 111.3.1, the mapping (3.4) is a homeomorphism of (T, onto
(X,p). Hence (3.4), which maps bijectively onto X(G), is a homeomorphism
we thus conclude that (3.4) is a
of (T3, r) onto (X(G), p). From
continuous mapping of (T5, r5) onto (X(G), p). The proof would be complete
if we had an inequality in the opposite direction between ri T5 and t5, to

demonstrate that these two metrics are topologically equivalent. Such an
inequality can be derived, for instance, by means of the right-hand inequality
(5.10) in 1.5.7. We content ourselves here with this remark, because we shall
and
in detail in the next section
study the relationships between t
(Theorem 4.7).
0
We remarked in 3.4 that if G is of the first kind,
is either a straight line
or a far from smooth quasicircie. For the corresponding quasisymmetric

functions fiR there is also a remarkable dichotomy. Suppose that G is a
finitely generated covering group of the first kind (e.g., the covering group of
a compact surface of genus > 1). If p is a trivial complex dilatation, then
R
is a singular function (Mosis the identity mapping. In all other cases
tow [1), Kuusalo [1]). The result holds for some more general groups G as
well, but it remains an open question whether it is true for all covering groups
of the first kind.
3.7. Unique Extremality and Tekhmller Metrics

of the upper halfLet Fh be the class of all quasiconfonnal
plane which agme with the quasisyinrnetric furnition h on the real axis. This
class was introduced in 1.5.7, where we noted that Fk alwayscontains an
With the help
extremal mapping with the smallest maximal dilatation in
of an example, it was shown that the extrernal inot necessarily unique.
Let us now make the additional assumption that hE X(G) for a Fuchsian
group G, i.e., that
hogoh1 = 9(g)1R,
where 8(g) is a Mbius transformation. We deflote by
whose functions I satisfy the condition fogo.f' =
(3.5)
the subclass of
By Theorem 3.4,
196
V. Teicbmller Spaces
is not empty, and again it contains an extremal mapping (cf.

2.2). If f1 and 12 are extremals in Fb and Fh(G), respectively, with complex
I = 1, 2, then
dilatations p, and maximal
the class
r(O,[p1]) =
z5(O,[p2])
(3.6)
The following observation establishes a connection between the extremals

the mapping O(g)of1 og' is also extremal in
in Fh and Fh(G). Along with
Fh. This can be verified immediately, in view of (3.5). Hence, if the extremal in
Fh is unique, it follows that
Ii ogof11 = O(g)IJf,
(3.7)
i.e., that
is not
Fh(G). Consequently, in this case wc see directly that
empty, without resorting to Theorem 3.4. Of course, is extremal in
and by (3.6), we have r(O,[p1]) = r5(O,[pl]).
We pointed out in 3.2 that there are cases in which
[p])> r(O, [p1). It
follows that whenever this occurs, the class Fk contains more than one extremal for h =
lift
4.
Teichmller Space and Schwarzian Derivatives
4.1. Schwarzian Derivatives and Quadratic Differentials

Again let S be a Riemann surface and G the covering
up of H' over S. If p
denotes a Beltrami differential for G, then by Theorem 3.1, the Teichrnller
space T5 can be characterized as the set of conformal mappings H. Let us
now form the Schwarzian derivative of
By Theorem 3.3, the mapping
o go
is a Mbius transformation for every g e G. It follows that
= S(f,ogof;1)of,JH =
(4.1)
We see that the
Schwarzian derivative of S111, is a quadratic

(a! for the group G acting on H. Its projection is a holomorphic quadratic
differential on the mirror image of the surface S. i.e., on H/G.

What we can actually deduce from (4.1) by reading it from left to right and
from right to left is the following result: Let f be'a conformal mapping of the
upper half-plane H. Then S1 is a quadratic differential for the group G if and
only iffo g of 1 agrees with a Mbius transformation in f(H) for every g e G.
Assume, in addition, that f is a mapping with
7 We then have a
third condition equivalent to the above two. For convenience of later reference we express all these conditions in a lemma.
Lemma 4.1. The following three conditions are equivalent:
10
20
Sf R is a quadratic differential for G;

og
agrees with a Mbius transformation
30 f11090(fP)l
agrees
ge
with a Mbius transformation on Flfor g.G.
197
PROOF. As we already remarked, the equivalence of 1 and 2 follows directly

from (4.1). If 2 holds, i.e., if
= w in
where w is a Mbius
transformation, then h =
H' H' is a Mbius transformation which coincides with o go (f")1 on R. Hence 3 follows from 2.
Conversely, assume that 3 holds, i.e., that
= h on
where
It is a Mbius transformation. Set w =

of
oh
the
in
and
tion, and so 2 follows from 3.
w is a Mbius transforma-
4.2. Spaces of Quadratic Differentials

Following the procedure in 111.4, we now introduce the counterpart of the
space Q. In what follows we regard G as acting in the upper half-plane II.
Let Q(G) be the space consisting of all functions qi holomorphic in the
upper half-plane H which are quadratic differentials for G and have a finite
hyperbolic sup norm:
<
h'II =
zeN
= x + iy. Like the space Q defined in 111.4.1, the space Q(G) has a natural
linear structure over the complex numbers.
The non-euclidean line element JdzI/(2y) is invariant under all conformal
self-mappings of H. It follows that, when e Q(G), y2 I
is invariant under
0. In the definition of the norm, we can therefore replace H by an arbitrary
Dirichiet region N c H of G:
=
(4.2)
z
ZEN
IfS is a compact surface, the closure of N lies in H. In this case it follows from
(4.2) that aU holomorphic quadratic differentials for G have finite norm.
If
isa subgroup of G2, we have the inclusion Q(02) Q(G1). In particular, all spaces Q(G) are subsets of the Banach space Q which corresponds to
the trivial group. From now on we write Q Q(l). Note that all spaces Q(G)
inherit a metric from Q(1); in 111.4 we introduced the notation q for this
dista1ice function.
Every Q(G) is a closed subspace of Q(l) and hence a Banach space. For
consider functions qx, e Q(G) which converge to q in Q(l). Given a g eG, we
then have
q,(g(z)), uniformly on every compact subset
of H. It follows that
= tim
= lim
o(z). Consequently, p e Q(G).
4.3. Schwarzian Derivatives of Univalent Functions

All points of Q(l), and hence of Q(G), are Schwarzian derivatives of functions
f meromorphic and locally injective in H. Let us consider the set
U(G)
{q = S1EQ(G)If univalent in H}.
198
Between
U(G) and U = U( 1) we have the simple relation
We just saw that Q(G) is closed in Q(1). In 111.4.4 we proved that U(1) is
closed in Q(1). It follows that U(G) is a closed subset of Q(1).
Since S1 H 6 whenever f is univalent, the remark in 4.1 preceding
Lemma 4.1 yields another characterization of U(G):
The set U(G) consists of the Schwarzian derivatives of those functions f

univalent in H for which
agrees with a Mbius transformation in
f(H) for every g e G. In other words, every f with S1 U(G) induces an
isomorphism of the group G QfltO the group G1 = {fogof1lgeG} of Mobius transformations acting on the domain f(H).
Let T(G) be the subset of U(G) consisting of the Schwarzian derivatives
plane
S1 U (G) of functions f which admit a quasiconformal extension to
with a complex dilatation that is a Beltrami differential for G. The set T(G) is
the image of the Teichmlier space T5 under the mapping
*
(4.3)
S1 IH
which fix 0, 1 and

This is clear: the normalization of the mappings
is unessential when we are considering Schwarzian derivatives.
In 111.4 we proved that (4.3) is a homeomorphism of the universal Teichmuller space (T,r) onto its image T(1) in (Q(1),q), and that T(l) is an open
subset of Q(1). We shall now start proving a sequence of theorems which, in
combination, assert thatthe restriction of(4.3) to T5 maps(T5,r5) homeomorphicalty onto T(G) and that T(G) is an open subset of Q(G). From ti T5
we conclude immediately that (4.3) maps (T5,
continuously into Q(G).
Let us briefly return to the MObius groups G1 induced by functions I with
S1 U(G). If
U(G)\T(G), then the G1-invariant domain f(H) is not a quasidisc. This follows from relation (4.4), which we shall establish ifl the next
subsection. The groups G1 with Sf lying on the boundary of T(G) are of

partic.ular interest and have been studied extensively by Maskit [1], AbikofT
[3], and others. (It is not known whether the inclusion
c (J(G)\T(G) is
proper in cases where G is not trivial; ci. 111.4.6.)
4.4. Connection between Teichmller Spaces and the

Universal Space
For the sets T(G) we have the natural inclusion T(G2)
T(G1) if G1 is a
subgroup of 62. Hence, all sets T(G) are subsets of T( 1).

The following fundamental result connects an arbitrary Teichmller space
in a simple manner with the universal Teichmlier space.
'Theorem 4.1. The Teichmller spaces satisfy the relation
T(G) = Q(G)
T(1).
(4.4).
199
4. Teichmullcr Space and Schwarzian Derivatives
PROOF. The inclusion T(G) Q(G) T(1) follows directly from the definitioris. We choose an arbitrary point S1 e Q(G) T(1) and prove that S, e T(G).
Let w be a conformal mapping of the lower half-plane H' onto the comSince
plement of the closure of f(H), normalized so that
(1(00))
S1e T(l),
the boundary of f(H)
is
a quasicircie. Hence, the function
of, defined on the real axis, is quasisymmetric; we may

h is normalized. Furthermore, for g e
h=
that
(4.5)
Since S1e U(G), the mapping Jo go

agrees with a Mbius
g1 in f(H). Then
ow, which maps H' onto itself, agrees with a Mbius
transformation 92 in H'. It follows from (4.5) that h induces an isomorphism
of G onto the group {g2lge G) of Mbius transformations acting on H'. In

other words he X(G).
Next we utilize Theorem 3.4. It follows that there is a quasiconformal
extension q of h to the lower half-plane which also fulfills the condition
w o is a quasiconformal extension
coo = g2 o for every g e G. Then f1
of f to the lower half. ,lane. For g E G we have
f1og =
g1
owoco
g1 of1.
This shows that f1 090

agrees with a Mbius transformation in
and it follows that S1e T(G).
(H'),
In the sixties, Bers posed the problem whether (4.4) is true. The above
proof is due to Lebto [2] who proved the conditional result that (4.4) holds
if and only iT Theorem 3.4 is true. After Theorem 3.4 was established, the
relation (4.4) thus followed immediately.
Theorem 4.1 allows important conclusions:
Theorem 4.2. The set T(G) is closed in T( 1).
PROOF. The relation (4.4) is equivalent tc T(G)

closed in Q(1), the theorem follows.
U(G) ri T(1). Since U(G) is
Theorem 4.3. The set T(G) is open in Q(G).
PROOF. This can be read from (4.4),since T(l) is open in Q(1).
Theorems 4.2 and 4.3 can be proved more directly, withobt the use of the
relation (4.4). Such alternate proofs will be given in subsections 4.6 and 4.7.
Bers [8] was the first to prove that T(G) is open; see also Earle [1].
Remark. Let S be the extended plane punctured at thrce points. Then S has
the half-plane as its universal covering surface. We noted in 2.7 that the
Teichmller space of S reduces to a single point. Consequently, T(G) consists
200
V. Tcichmller Spaces
of zero only. Since T(G) is an open subset of the linear space Q(G), we deduce
via'the Teichmiler theory that in..shis case, Q(G) does not contain points
other than the origin
e
4.5. Distance to the Boundary

By Theorem 4.3, the mapping (4.3) carries a neighborhood of the origin of T5
onto a ball of Q(G) centered at the orign, but we can say more.
Theorem 4.4. The ball
B(0,2) =
<2)
lies in T(G).
PROOF. In 111.4.3 we remarked that (g' e Q( 1)111
<2} lies in T(1) (Theorem
11.5.1). Hence, the theorem follows immediately from (4.4).
There is another more direct way to show that B(0, 2) is contained in

T(G). Let cp e B(0, 2) and set
= 2y2
Since p(g(z))g'(z)2 = q4z) and
= Ig'(z)VImg(z), we see that p is a Beltrami differential for G. Hence
[p3 e T5. From what was said in 111.4.3 we know that the mapping
(4.6)
is the inverse of [p3 ' = S1 H in B(0, 2).

Since
6fl([p],0) (formula (111.4.1)), we conclude that the ball
I
{[pJeTsIP('pI,O) < 1/3)
(4.7)
is contained in the preimage of B(0, 2).

Theorem 11.5.1 says that the largest ball in T(1) centered at the origin has
the radius 2. The Schwarzian derivative of the logarithm is a boundary point
of T(l)with distance 2 from 0. In the upper hall-plane the logarithm is cornwith the group consisting of all Mbius transformations of the form
z ' az or z a/z, where a is a positive real number. It follows that if G is a
properly discontinuous subgroup of this group, then B(0, 2) is the largest ball
in T(G) with center at 0. For instance, this is the case in the Teichmller space
of an annulus (cf. IV.4.3). For an arbitrary G, the determination of the largest
ball in T(G) centered at 0 remains an open problem.
We shall use the mapping (4.6) in section 5 in studying the complex analytic structure of
Here we draw the following conclusion.
Theorem 4.5. Every point of the Teichmller space T3 can he represented by a
real analytic Beltrami diffirential and by a real analytic quasiconformal mapping.
PROOF. Let a point [p3 = p e be given. Suppose first that p can be represented by a quasiconformal mapping whose maximal dilatation is <2. Then
201
4., TeiclunUller Space anSI Schwarzian Derivatives
p lies in the set (4.7), and so p can be represented by z -. 2y2q(i), which is

a real analytic complex dilatation. We also have an explicit expression for the
corresponding mapping f" (cf. 11.5.1 and 11.5.2) from which it becomes apparent that is real analytic.
The general case is handled by induction. Assuming that the theorem is
true jf
0) <r, we show that it holds if
0) <2r. Let fP be an extremal
ofM2, where p2(z) is the
for the point p, with rs(p,O) < 2r. We write fTM =
middle point of the line segment from 0 to
in the non-euclidean metric of
the unit disc (ci. Theorem 111.2.2). Then
is a Beltrami differential for G and
a Beltrami differential for
Moreover,
0) <r and r5.([/1j], 0) <.
r with 5' = if/0U2. From this the theorem follows.
0
Theorem 4.4 can be generalized. Every point [p] e T5 determines uniquely
the quasidisc
Puttingtogether a large part of our previous ana=
lysis, we obtain a lower bound for the distance to the boundary of the point
s,1 of T(GJ in terms of the inner radius of univalence (defined in 111.5.1) of
Theorem 4.6. For every sp
T(G), the ball
<
B(s,1,0,(A,1)) =
is contained in T(G).
In 111.5.3 we proved that
<
Consequently, the theorem follows immediately from (4.4).
We recall that in T( 1), the inner radius
to the boundary (Theorem 111.5.1).
lies in T(1).
(A,1) is precisely the distance from
4.6. Equivalence of Metrics

The topological equivalence of the metrics r5 and rj can be proved with
the aid of Schwarzian derivatives or of quasisymmetric functions. Neither
method goes to the heart of the matter, but both give the desired result easily.
The method based on tht use of Schwarzian derivatives produces better
constants. In fact, we can prove that the metrics are even uniformly equivalent, i.e., the identity mapping (T5, 'rj T5)
(T5, t5) and its inverse are uniformly
continuous.
Theorem 4.7. The metric of the Teichmller space
is unjformly equivalent to the metric ti T5 induced on T5 by the metric of the universal Teichmller
space. For every poiflt p e
r(p,q)
3.
(4.8)
V. TeichmUller Spaces
302
More generally, on every bounded set A c T5,
3(1 + diaA)r(p,q),
(4.9)
where dia A denotes the diameter of A in the ta-metric.
and jie p. We know that
PROOF. Let p e
On the other hand, it follows from Theorem 4.4 and formula (4.6) that
Hence,
(4.10)
3f3(p,O).
In order to generalize this estimate for an arbitrary pair of points of T5, we

fIx p = [j.t] T5 and consider the mapping
defined by
=
The function &,, induces a well-defined mapping
[v]
which is an
(4.11)
bijection of the universal Teichmller space T onto
itself. It maps T5 dto the Teichmller space

where S" = H'/Gu, and it is
an isometry of (T5,
onto
r5). The last assertion follows directly from
the definition of the Teichmllcr metric.
The mapping (4.1 1) takes the points p =
and q = [vJ ofT5 to the points
0 and [A] = ;([v]) of
Hence, by (4.10),
< 3/3(0,[A]) =
135(p,q) =
313(p,q).
Since trivially 13 135, we have established the double inequality
fl(p,q) < 135(p,q) 313(p,q)
(4.12)
for all points p and q of T5.
The inequalities (4.12) show that the metrics PITs and
are uniformly
equivalent. Since
fi=tanht,
(4.13)
thc Teichmller metrics nT5 and rs are also uniformly
it
equivalent.
From (4.12) and (4.13) we obtain
1
1+3t
3t
13t
l3r
r5log
if
<
1/3.
This yields inequality (4.8). Also
203
(4.14)
if t
a < 1/3.
Let A be a set in T5 with a finite diameter dia A in the Trmetric, and p and
that (4.14) holds. If t(p, q)> a,
q points of A. If t(p, q) a, we just
then
.
dia A/3(1 + dia A) we obtain (4.9).
r5) is complete. If the

proved in 3.3 that the Teichmller space
result is combined with Theorem 4.7, we conclude that
T
is true of their homeomorphic images, we have reproved
We
Theorem 4.2: T(G) is a closed subset of T(l).
4.7. Bers Imbedding

Theorem 4.7 makes it possible to generalize Theorem 111.4.1 immediately
(Bers [8]).
[p]SJ1H
is a homeomorphism of(T5,
(4.15)
onto (T(G), q).
PRooF. By Theorem 111.4.1, this mapping is a homeomorphism of(Ts,rITs)

onto T(G). By Theorem 4.7, the metrics t5 and ti are equivalent, and the
theorem follows.
0
In view of Theorems 4.3 and 4.8, we are again justified in calling the
mapping (4.15) the Bers imbedding of the Teichmller space.
In 4.5 we proved that B(O, 2) c T(G) without utilizing relation (4.4). We
shall now show, without resorting to (4.4), that every point of
has an
open neighborhood in Q(G) which is contained in T(G).
To prove this, let us consider in T(1) the mapping i/i defmed by 'I'(s,) =
If). is the Bers imbedding of the universal Teichmiiller space onto T(1),
then t/i =
(Here
and are as in 4.6.) It follows that is a
homeomorphism of T(1) onto itself. Furthermore, maps T(G) onto T(G"),
and Q(G) T(1) is mapped onto
n T(1) (Lemma 4.1).
Let V be an open ball in
centered at the origin which is contained in
Write V = Q(GsL)n
V0, where V0 is a neighborhood of the origin
in T(1). Since T(1) is open in Q(1), the preimage ,/r'(V) = Q(G) i,lr'(V0) is a
204
neighborhood of
in Q(G). This neighborhood is contained in i,Iii(T(GU)) =
T(G).
We have proved without making use of(4.4) that T(G) is closed in ii!) and
open in Q(G). These results imply that T(G) is open and closed in
T(1).
Since T(G) is connected, it follows that T(G) is the component of Q(G) T(l)
containing the origin. This result can be used instead of (4.4) in proving
Theorem 4.6.
Remark. In 111.4.5 we proved that T(1) = mt U(1). It follows that T(l) =

mid 7(1), i.e., every neighborhood of every boundary point of T(l) contains
points which are in the complement of the closure of
In the general case it is not known whether T(G) agrees with the interior of
U(G). However, it is Irue that
T(G) = mid T(G).
(4.16)
For finitely generated groups G of the first kind, this was proved by Abikoff
[3] (who resorted to an unpublished result of Thurston). Quite recently.

[I]
(4.16) was established by Bers [13) in the general case. In a note,
has announced the result that T(G) is the component of mt U(G) containing
the origin. By aid of Luravlev s result, Shiga [1] proved that if G is finitely
generated and of the first kind, then indeed T(G) = mt U(G).
4.8. Quasiconformal Extensions Compatible with a Group

We conclude this section by showing that for suitable reflections, the Ahifors
extension of Theorem 11.4.1 is compatible with the action of a group.
Given a point S1 of T(G) and hence a quasicircie
we consider quasiconformal reflections A = of of;' for all
in the equivalence class. We
show that there always exists a such that 1 is Lipschitz-continuous.
Since is compatible with the group G, i.e., og
is a Mdbius transis compatible
formation for every ge G. it follows from Lemma 4.1 that
with G. We construct a quasiconformal mapping i/i: H' H' with boundary
values
by means of the DouadyEarle method (see 3.6). Then i/i is
G-compatible.
Let us.consider
G-compatiole, because
to be the extension of
in H'. It agrees with
on Il', and it
is
and i/i are G-compatible. We take f,, o(JM)_i o
to the lwer half-plane.

Now cli is a diffeomorphism and Lipschitz-continuous in the hyperbolic
metric of H'. We conclude exactly as in Lemma 1.6.4 that with our choice of
is Lipschitz-continuous and, with the
the reflection A =
exception of
continuously differentiable. In 3.5 we proved that
(4.17)
for every element of the deformed group GM =

g OfU' J9 E G}.
=
compatible with GM. Following
Letf be a univalent function in AM =
S. Complex Structures on Teichmller Spaces
205
the proof of Theorem 11.4.1, we use the representation f = w1 /w2, where

w2 is a normalized pair of solutions of the differential equation w"
and set for z e AM,
We assume that f is holomorphic on
w1,
w(z)
W2(Z)
+ (2(z) z)w(z)
+ (2(z)
By Theorem 11.4.1, there is a positive constant c, which depends only on

[p3, such that if uS1 II <c, then wo 2 is a quasiconformal extension off (A
modification with the help of a Mbius transformation is required, because

is unbounded. The bound e need not be the same as in Theorem 11.4.1,
because 2 is constructed differently.) Assuming that ITS,lI <e, we shall prove,
as a supplement to Theorem 11.4.1:
The quasiconformal
PROOF.
extension w o
2 off is
compatible with the group
The result follows by direct computation. First of all (cf. 111.5.4),
A(z)
(2(z) z)2S1(z)
282(z)
Making use of the identity (g(z1) g(22))2 = g'(z1)g'(z2)(z1 22)2, which

holds for all Mbius transformations g, and because of (4.17), we obtain
(2(gM(z))
z)2.
Furthermore,
=
and by our hypothesis,
These formulas yield
2aA(z)
From this and (4.17) it follows that

(4.18)
=
By (4.17),
Combined with (4.18), this shows that
5.
is
compatible with GM.
Complex Structures on TeichmUller Spaces
5.1. Holomorphic Functions in Banach Spaces

With the aid of the Bers imbedding we can introduce a natural complex
analytic structure into the Tcichrnllcr space
The Tcichmller space thus
V. TcichmUfler Spaces
206
becomes a complex analytic Banach manifold, a generalization of the notion

of a complex analytic n-manifold defined in IV.l.2.
First of all, we shall recall the definition of holomorphic functions in

Banach spaces. Let E and F be Banach spaces over the complex numbers,
and U c E an open set. A function f: U . F has a derivative at a point

x0 e U ii there exists a continuous complex linear mapping Df(x0): . F
such that
+ h) f(x0)
Df(xo)(h)IIF
h-O
The mapping Df(x0) is called the derivative of f at x0. A function f: U . F

which has a derivative at every point of U is said to be holomorphic in U.
The composition of holomorphic functions is holomorphic where defined.

A holomorphic function f: U . f(U) is biholomorphic if it has a holomorphic
inverse. In the case E Ctm, F = CA, the above definition coincides with the
usual notion of holomorphic functions.
In order to get a connection with ordinary complex-valued analytic func-
tions, we introduce the dual P of F; The set F consists of all continuous

complex linear mappings of F into C. The norm
Jx*
makes
P a Banach space. A set A c
ItxIi,
l}
P is called total if
= 0 for every
-
There are two characterizations which, taken together, make it possible to
consider only complex-valued functions of a complex variable when it comes

to checking whether a mapping between Banach spaces is holomorphic.
Lemma 5.1. A function f: U . F is holomorphic
of the following two conditions:
and only
it satisfies one
(i) For every XE U and e E E, the function w , f(x + we) is a holomorphic

function on an open neighborhood of the origin in C with values in F.
(ii) The function f: U . F is continuous and there exists a total subset A of the
dual P such that, for every e A, the function of: U . C is holomorphic.
Conditions (i) and (ii) are given in Bourbaki [I], 3.3.!.
5.2. Banach Manifolds

A complex Banach manifold M is a Hausdorif space with an open covering
of sets each of which is homeomorphic to an open subset of a complex

space (not necessarily the same for the open sets of M). Suppose M
has an atlas in which all parameter transformations are biholomorphic. A
maximal atlas with this property is called a complex (analytic) structure on
S. Conij$ex Structures on TeichmUllcr
207
M. A manifold M with a complex analytic structure is called a compiex
analytic Banach mamfold.

Theorems 4.8 and 4.3 assert that the Teichmller space T3 is actually
bally bomeomorphic to an open set in the complex Banach space

which consists of the quadratic differentials with finite norm for the covcrh*
group G. Hence, by using this result, we could make the Teichmller space a
complex analytic Banach manifold.
We shall show that, in fact, a natural complex structure for T5 is obtained
in this manner. However, the required verifications are easiest to carry out
with the results at our disposal if we introduce the complex structure a little
differently, using local parameter mappings.
A complex analytic structure can be given to a Teiclunllcr space in several
ways. This is particularly true in the case of compact Riemanji surfaces. On
the other hand, the various approaches lead to isomorphic structures, so that
we are free to speak of a canonical complex structure on a Teichmller space.
Here we shall be dealing with the general case, excluding only those Ricmann
surfaces which do not have a disc as a universal covering surface. That we
arrive at a natural complex structure is seen from Theorems 5.25.6.
In section 9 we shall construct the "Teichmuller imbedding" which shows
that the TeichmUller space of a Riemann surlice of genus p is homeomorpbic
to C3"3. However, the complex structure the Tcichmller spaces inherit.

through this imbedding is not a natural one: Given two Riemann
surfaces Sand S of genus p. the bijective isometrybetween and T5. induced
by a quasiconformal mapping of S onto S' is usually not biholomorphic with
respect to these structures. Tcichmller was aware of this state of affairs, but
from
in one of his last papers.(Teichmuller [3]), he claimed to have proved the

existence of the "right" complex structure. This paper is difficult to read, and
today Teichmller's reasoning on this point is not regarded as convincing.
The first complete proof for the pxistence of the complex structure in
Teichmller spaces of compact surfaces is due to Ahlfors [21; see also Bers
[2]. Subsequently Bers ([7], 18]) introduced complex structure into an arbitrary TeichmUller space by means of the mapping p + S1.
5.3. A
Holomorphic Mapping between Banach Spaces
The introduction of the complex structure on the Teichmller space
is
based. on the following result.
Theorem 5.1. The function

p . A(p) = S1111.
(5.1)
which maps the open unit hail B(G) of the space of measurable ( I, I)for G into the space Q(G) qf holomorphic quadratic
for Q, is holonwrphic.
V. Tcichmulkr Spuccs
208
Ppow. We have already seen that Q(G) is a Banach space. The ball B(G) is
an open subset of the Banach space LIG) of measurable(I, 1)-differentials
v
Fix
B(G).
for G with finite
is a total
For
we set
and
= q(z). Then A =
set in the dual of Q(G). We apply condition (ii) of Lemma 5.1 to the function
. s(w)
(5.2)
The set U in condition (ii) is now the neighborhood {wIIwI <(1 Ii4uL)/
of the origin in the complex plane. Then p + wv e B(G). Furthermore,
II v II
let F = Q(G) and
By Corollary 11.3.1 and the Remark following it, the function
=;.
CS =
holomorphic in U for every z E H. By formula (111.4.4), the function s is

continuous in U. Hence, by condition (ii) of Lemma 5.1, the function (5.2) is
holomorphic in U. Using this fact we conclude from condition (i) of Lemma
is
5.1
that (5.1) is holomorphic in B(G).
though the definition of the complex structure on T5 with the aid of

the holomorphic mapping (5.1) requires a number of auxiliary mappings, as
illustrated by the diagram in Fig. 14, the idea is quite simple. The ball B(O, 2)
of Q(G) plays a distinguished role, because it follows from what was said in
4.5 that (5.1) has a holomorphic section there. In the preimage of B(O, 2) in T5
we take [vJ 9 S,IH as a local parameter. Near an arbitrary point [pJ of T5 a
local parameter is obtained if we first map isometrically onto
in such a
way that [pJ is movd to the origin. An easy verification shows that the
parameter transformations
biholomorphic. In 5.4 and 5.5 we shall explain all this in detail.
Even
5.4. An Atlas on the Teichmller Space

Let us now introduce the auxiliary mappings which are needed in the definition of the complex analytic structure on T5. For a given
B(G), we
again write GU =
As in 4.6, we consider the mapping
B(G) -.
B(G'9, defined by
f&..(v)
= fv0(fP)1,
or, in more explicit terms, by
=
The function &,, maps B(G) bijectively onto B(GM). (Application of
(5.3)
sim-
ply means transforming the Riemann surface S quasiconformally to the SUr-
209
Structures on TeichrnHcr Spaces
5.
B(G)
Figure 14
face S'1 = If'/G'1.) From (5.3) and the definition in 5.1 it follows that
bihol om orphic.
In 4.6 we noticed that the induced mapping ;, defined by
is a bijective isometry of onto
we write
For p 8(G), v
is
(5.4)
=
By Theorem 5.1, this mapping of
Q(G'1) is holomorphic.
2)
From the proof of Theorem 4.4 we know that in the ball
<2}, the
Q(G"fl
(5.4) has a section
2) .
defined by
zEH.
(5.5)
From (5.5) and the definition in 5.1 we see that or,, is holomorphic.
/
Let it denote the canonical projection of B(G) onto
the;
and A and
Bers imbeddings of and
respectively, into Q(G) andQ(GM). The commutative diagram in Fig. 14 illustrates the mappings introduced here.
The collection
=
an open covering of
homeomorphism
is
(5.6)
In fact,
is the preimage of B,,(0, 2) under the

(5.7)
of
onto T(G").
5.5. Complex Analytic Structure.,
We shall now prove that the restriction mappings
define a complex
analytic structure for the Teichmller space T5 of the Riemann surface S and
that this structure has the expected good properties. The proof does not
cover the case of a torus; the complex structure of the Teichmller space of a
torus will be introduced in section 6.
210
V. TeichmURer Spaces
Theorem 5.2. The atlas

(5.8)
The Bers
defines a complex analytic structure on the Teichmller space
holomorphic
with
to this
imbedding
into
is
respect
of
T5
Q(G)
'
structure.
PROOF. Assuming that
and are defined by (5.6) and (5.7), we choose two
is not empty. In
elements
and p2 of B(G) such that Vi,,
we have
=
All mappings on the right-hand side are holomorphic, as we have seen.
o
is holoConsequently, as a composition of holomorphic functions,
morphic. By changing the roles of p1 and P2 we conclude that
is
biholomorphic. Hence, (5.8) defines a complex analytic structure for T5.

It is not difficult to see that the complex structure we obtained by means
of the atlas (5.8) is independent of the representation H'/G we used for the
Riemann surface S. Theorem 5.5 expresses an even stronger result.
In order to complete the proof of the theorem, we still have to show that
the Bers imbedding 2: T5 Q(G) is holomorphic, i.e., that 1
is holomorphic in
2). Now
=
Ao'
oak.
Since all mappings on the right are holomorphic, their composition lo h' is
holomorphic.
We shall now establish further results (Theorems 5.3-- 5.6) which show that
(5.8) defines a natural structure.
Theorem 53. The canonical projection
it: B(G)
is
T5
holomorphic, and it has local holomorphic sections everywhere in T5.
PROOF. First of all, we have
=
and are holomorphic, it follows that it is holomorphic.
Next, let us consider the mapping
Since
of into B(G). All functions on the right-hand side are holomorphic. Therefore, their composition i/ia is holomorphic. From the definitions we infer that
ito
Consequently.
is
= identity mapping of
the desired section.
(5.9)
211
5. Comple, Structures on Teichmller Spaces
that it is an open mapping. We remark

= we
that Theorem 5.3 determines the complex analytic structure of uniquely.
Since
Theorem 5.4. The Bers imbedding
T(G) is biholomorphic.
T5
PROOF. Suppose first that Q(G) is finite dimensional. (By IV.5.5, this is the
case ifS is compact; cf. also 9.7.) We then conclude directly from Theorem 5.2
that 1t is biholomorphic using the theorem by which a holomorphic bijection
is always biholomorphic in finite dimensional manifolds (Narasimban [1],

p. 86).
In the general case, we fix a point
E T(G) and consider Schwarzians

has a small Schwarzian derivative in the quasidisc fM(H). One proves that f has a quasiconhorinal extension
whose complex dilatation is in B(GM) and depends holomorphically on s.,
(Bers [9], Theorem 6; cf. the remark at the end of 11.5.1). The conclusion is
that A has a local holomorphic section at
Since A = o it, we deduce from
Theorem 5.3 that is biholomorphic.
0
s,e T(G)lying close to
Then I =
5.6. Complex Structure under Quasiconformal Mappings

Theorem 2.6 states that the Teichmller spaces of quasiconlormally equivalent Riemann surfaces are isomorphic in the sense that there exists a
bijective isometry between the Teichmller spaces. We can now strengthen
this result and show that such an isometry can be chosen to be biholomorphic.
Theorem 53. Quasiconformally equivalent Riemann surfaces have isometrically

and biholomorphically isomorphic Teichmller spaces.
PRooF. Let S = H'/G and S' H'/G' be quasiconformally equivalent Riemann surfaces. We consider a lift of a quasiconformal mapping of S onto S'
to a self-mapping of the lower half-plane. There is no loss of generality in
assuming that the lift is of the form fJL. We have
B(G).
The mapping
induced by p is now a biholomorphic mapping of
onto B(G'). It induces the mapping of T5 onto T5.. The theorem follows
when we prove that ; is biholomorphic.
If
is the canonical projection, we have
B(G') '
(5.10)
Consider an arbitrary element vEB(G). in view of(5.9) and (5.10), we have in
Again, on the right all functions are holomorphic, and so is bolomorphic.
By changing the roles of S and 5' we conclude that
is biholomorphic. 0
212
V. TeichmUfler Spaces
To be quite precise, we have not yet proved Theorem 5.5 for tori, which
have the complex plane rather than the half-plane as universal covering
surface. In the next section we shall show that the TeichmUller space of every
torus is biholomorphically equivalent to the upper half-plane.
Theorem 5.5 can be applied to the modular group Mod(S) of T5, which was
introduced in 2.7.
Theorem 5.6. The elements of the modular group Mod(S) are biholomorphic
automorphisms of the Teichmller space T5.
PRooF. By Theorem 5.5, a quasiconformal mapping between the Riemann

surfaces S and S' induces a biholomorphic isoniorphism T5 *
The ele-
ments of the modular group are such isomorphisms induced by..-quasiconformal self-mappings of S.
D
We proved in 1.5 that two compact Riemann surfaces are -quasiconformally equivalent as soon as they are homeomorphic. Moreover, we learned
in IV.5.2 that they are homeomorphic if and only if t)iey have the sam genus.
Therefore, in view of Theorem 5.5, the genus
determines the TetchThe abstract
muller space of a compact
space corresponding to surface* of genus p is denoted by T,,. Starting from a specific
Riemann surface S, as we have done, means fixing the origin in 1;.
In IV.5.5 we saw that the space of holomorphic quadratic differentials of a
compact Riemann surface..of genus p> 1 has finite dimension 3p 3. Therefore. Q(G) can be identiFied with
T, becomes a complex analytic
(3p 3)-manifold. We shall consider this question in greater detail in section
9 by using another imbedding of T5 into Q(G). In section 6 we shall prove that
'T1 is a complex 1-manifold.
6. Teichrniiller Space of a Torus

6.1. Covering Group of a Torus
In sections 354e considered Riemann surfaces having the half-plane as a
universal covering surface. Let us now assume that S is a Riemann surface
which has the complex plane C as its universal covering surface. If the
covering group G of C over S is trivial or cyclic, then all quasiconformal
images of S are conformally eqpivalent (ef. IV.4. 1). Leaving aside these cases,
which are uninteresting from the point of view of the theory of Teichrnller
spaces. we assume that S is a torus, i.e., a compact surface of genus 1. By the
remark at the end of 2.6, the Teichmller spaces of different tori are all
isom&rically bijective.
213
6. Teichmfler Space of a Torus
The covering group G of C over a torus consists of translations
are complex numbers, Im(co1/co2) 0, and m and n run

where w1 and
through all integers 1. We call the pair (Wi, W2) a base of G; the base is said
to be normalized if Im(co1/w2) > 0.
is a base.of G jf
L*mma 6.1. Let (w1,co2) be a base of G. Then

only
and
if
= aw1 + bw2,
=cco1 + dw2,
(6.1)
where a, b, c, d are integers and
adbc=1.
(6.2)
is normalized if and only
if(co1,co2) is normalized, then
ad
bc
1.
PROOF. The validity of (6.1) with integral coefficients is clearly a necessary

condition. it becomes sufficient if (6.1) can be solved with respect to co1 and
w2 so that
and (02 are linear combinations of
and
with coefficients
in. Z. This occurs if and only if (6.2) holds.
From (6.1) it follows that
adbc
(6.3)
\W2)
\W'2J
Assuming that (6.2) is true we see that Im(w1/co2) and

taneously positive if and only if ad bc = 1.
are simul-
Consider the elliptic modular group f which consists of the restrictions to

H of all Mbius transformations
z
where a, b, c, d are
az+b
_____
cz
+d
real integers and ad
bc
1. By (6.3) the group F acts
on H.
The group r contains elliptic transformations. These are of two types.

Either they are conjugates of the mapping z -+ lJz, which has the fixed
point i in H. These transformations are of period 2. Or they are conjugates of
in H, and are then
z - 1/(z + I), which has the fixed point 1/2 +
of period 3. Even though r is not fixed point free, it is not difficult to prove
that F acts properly discontinuously on H. (For more details about the

properties of F we refer to Lehner [1], pp. 87, 99, and 139.) By Theorem
IV.3.2 and the remark following it, the quotient H/T is a Riemann surface.
Let S = C/Gibe a torus and (w1,w2)a normalized base of G. We conclude
from
is a pair of non-zero complex numbers, then
6:k that
is a hormalized base of G for a
0 if and only if
is
214
V. Teichmller
equivalent to to1 /0)2 under the modular group r. Hence every torus can be
represented as a point of the Riemann surface H/r.
6.2. Generation of Group Isomorphisms

Theorem 1.3 says that if S is a Riemann surface which admits a disc as its
universal covering surface and for which the covering group is not elementary, then the only conformal deformation of S onto itself is the identity
mapping. For tori the situation is quite different:
Lemma 6.2. Let S be a torus and p and q arbitrary points of S. Then there is a
conformal mapping f: S , S homotopic to (he identity such that f(p) = q.
it: C 'S = C/G be the canonical projection and zir'{p}, WE
{q}. A translation commutes with every g E G. Therefore, the mapping
.
+ t(w z) can be projected to a conformal mapping j: S S for every
t, 0 t I. As t varies from 0 to 1, we obtain a homotopy from the identity
mapping to f1 = f, and f(p) q.
0
PROOF. Let
Lemma 6.2 tells us that in studying the Riemann space R5 and the Teichof a torus we can restrict ourselves to mappings
S'
S
which are normalized by a condition p(p) = p'. pES, p'ES'. If it: C iS and
it': C ' S' are the canonical projections and we take p = ir(O). p' = n'(O), then
C with the property f(O) = 0. We call such a lift f
4, has a unique lift f: C
mUller space
normalized. A normalized f induces an isomorphism of G onto G' under

which the transformation z , z + mw1 + nw2 maps to z ' z + mf(co1) +
nf(w2).
In the case of tori it is easy to show that every isomorphism between G and
G' can be generated in this manner.
Theorem 6.1. Let S = C/G and 5' = C/G' be tori and 0: G i G' an isomorphism. Then there is a homeomorphism of S onto S' which induces 0.
PROOF.
Let (w1,w2) be a base of G and suppose that (w1,a2) ,
('l,'2)
under 0. Consider the affine transformation a which fixes 0 and maps
to
cot, i = 1, 2. Then a determines 0, and it projects to a homeomorphism of S
onto S'.
0
Let ip: S ' S' be a homeomorphism with a normalized lift f such that
f(co1) = w, i = 1, 2. If h =
then h(z + w3 = h(z) + w. From this we
conclude that h is sense-preserving (Theorem IV.3.5). Hence f and a are
simultaneously sense-preserving.
Let r = a,1/ai2, r' =
If r'
I, then
a(z) = 2(z + Pr).
(6.4)
6. Teichrnller Space of a Torus
215
Direct calculation shows that

(6.5)
Suppose that (w1, COi) is a normalized base of G, ie., that Im t > 0. By (63),
we then have lar' > 0 if and only if

< 1 is equivalent to a
< 1. But
being
It follows that
is sense-preserving if and only if
'Im(f(o1)/f(w2))> 0. 11 = then qi is sense-reversing and Im(f(w2)/
<0.
If
< 1, then (6.4) defines a quasiconformal mapping. Since the projec-
tion of a quasiconformal a is quasiconformal, it follows that all tori are

equivalent. We also conclude from the proof of Theorem
6.1, by use o( Theorem IV.3.5, that to every sense-preserving homeomorphism p between two tori S and S' there is a quasiconformal mapping of S
onto S' which is homotopic to 4p. (This is a new proof of Theorem 1.5 for
tori.)
6.3. Conformal Equivalence of Tori

Let S = C/G and S' = C/G' be two tori and
a base of G. We first
show that S and S' are conformally equivalent if and only if there is a
complex number A
0 such that (Ao1 , Aco2) is a base of G'.
In fact, if S and S' are conformally equivalent, then by Lemma 6.2 there
is a conformal map of S onto S' whose lift f: C -. C is normalized. Since f
is a conformal self-mapping of C with f(O) = 0, we have f(z) = Az. Then
and so
is a base of G'. Conversely, if (Aw1,Aco2) is a
base of G', then the projection of z ' Az is a conformal mapping of S onto S'.
Theorem 6.2. Let S C/G and S' = C/G' be tori and (co1,co2) and
normalized bases of G and G'. Then S and S' are conformaUy equivalent if and
only the points w1/co2 and o11/w'2 are equivalent under the elliptic modular
group.
PROOF. We just showed that S and are conformally equivalent if and only
if there is a A 0 such that 4Aw1,Aw2) is a base of G'. From what we
at the end of 6.1 it follows that this is the case if and only if the points w1 /w2
and
are equivalent under the elliptic modular group.
Theorem 6.2 provides a model for the Riemann space R5 of tori. We

conclude from it that R5 can be mapped bijectively onto the quotient of the
upper half-plane H by the modular group r. The mapping x: R5 -. H/I' is
obtained if we fix a normalized base (cot, w2) for the covering group of C over
Sand set
= [f(co1)/f(o2)] for peR5,fEp.
Repeated application of the reflection principle shows the existence of a
216
V. TeichmUHer Spaces
"modular" function J, which is holomorphic in H, maps II onto C, and is

autornorphic with respect to the group IT. By using J we can identify H/IT
with the complex plane (Lehner [1], pp. 45).
Let f be a conformal self-mapping of S. The lift of f to C is then of the form
2: -i- constant. If A = I or A
z
1, then conversely, the projection of
z
+
Constant
IS
conformal
self-mapping
of S. These are the trivial
a
conformal self-mappings of a torus.

In certain exceptional cases, a torus admits other conformal self-mappings.
We deduce from the preceding' considerations that z --' Az projects to a conformal self-mapping of S for some A
I if and only if w1/c02 h(a1/w2)
for a transformation he IT different from the identity mapping. In other

words, w1/w2 must be a fixed point of an elliptic transformation heIT. As
stated before, there are two possibilities. First, h is a transformation of period
2, i.e., a conjugate of z + liz. Then (w1,w2) can be chosen such that the
lundamental parallelogram with the vertices 0, cot,
+ w2, w2 is a square,
and A = i. Second, h is of period 3, i.e., a conjugate of z
1/(z + 1). In
this case there is a fundamental parallelogram which again has sides of equal
length hut the angles are 7t/3 and 2it/3, and A = 1/2
These are the
only cases in which a torus admits non-trivial conformal self-mappings.
The rest of this section is concerned with showing that the TeichmUller
space of a torus is isomorphic to the hyperbolic upper half-plane (or to the
unit disc). It follows, in particular, that the Teichmller space of a torus is
branched over the Riemann space at points which correspond to the fixed
points of F or, what is the same, which correspond to symmetric tori admitting non-trivial conformal self-mappings.
6.4. Extremal Mappings of Tori

We obtain a key to the properties of the Teichmller space oC a torus by
studying mappings which are extremal in a homotopy class.
following
basic result is due to Teichmller ([1], p. 31).
Theorem 6.3. In each homotopy class of sense-preserving homeomorphisms

between tori there is a mapping whose is are affine and which is extre,nal, i.e.,
which has smallest maximal dilatation. Having affine lifts or being extretnal
determines the mapping uniquely up to conformal mappings homotopic to the
identity.
PROOF. Let
S = C/G and S' = C/G' be tori, and consider mappings in a given
horn otopy class of sense-preserving hoineomorphisms of S onto S'. By Lemma

6.2 and the fact that conformal self-mappings of C are afline, we may restrict
ourselves to mappings with normalized lifts. Let F denote the class of these
normalized lifts.
217
6. Teichmller Space of a Torus
Fix a normalized base (w1 , co2) of G. There exists a normalized base

of G' such that F is the class of seif-homeomorphisms .f of C
satisfying f(O) = 0 and
f(z +
for all z e C and m,
mw1
+ nco2) = f(z) +
+ na4
(6.6)
7L. This follows from Theorem IV.3.5, in view of the fact
that the identity is the only inner automorphism of G'. Clearly, there is a
unique affine mapping w in F.
Let feF be K-quasiconformal, and set fk(z) = f(kz)/k, k = 1, 2, .... Then

every fk F is K-quasiconfoEmal. From (6.6) we conclude that for k cc, the
mappings fk converge to w, uniformly in the euclidean metric. By Theorem
1.2.2, the maximal dilatation of w is at most K. It follows that the projection
of w is extremal in the given homotopy class.
We now prove that there are no other extremals in this homotopy class.
There is no loss of generality in assuming that w is a stretching in the direction of the real axis, or more precisely, that w(x + ly) = Kx + iy, K 1. In
fact, this normalization can he obtained by suitable conformal self-mappings
of C, which do not change the maximal dilatation. Note that K is the maximal dilatation of w.
Let f be an extremal mapping competing with w. We prove that f =
w.
Sincef w has the periods w1 and co2, there is a positive number L such that
If(z)w(z)IL
(6.7)
for every z e C. Since f is absolutely continuous on lines, we infer from (6.7)
that for every r >
0,
flfx(x + iy)Idx
for almost all
Jfx(x +
[0, rJ. If Q = {(x, y)IO x <r, 0
JJQ
Kr 2L
y
r}, it follows that
Kr2 2Lr.
As r , cc, the number of period parallelograms meeting Q is of the order

of magnitude r2(1 + o(1))/m(P), where m(P) is the area of the period parallelogram P. Hence,
Kr2
cc
2Lr
we obtain
Km(P)
As an extremal,f is K-quasiconformal. Therefore, lf,J2 KJ a.e.,'where J
218
is
the Jacobian of f (cf. 1.3.4). Application of Schwarz's inequality yields
K2(m(P))2 m(P) J
dx dy Km(P)m(f(P)).
(6.8)
Because f(P) and w(P) are fundamental domains of G' with boundaries of
measure zero, m(f(P)) = m(w(P)) = Km(P). We see that equality holds everywhere in (6.8), and so
= KJ(z)
almost everywhere.
We write
+
of + all =
t3j J = tail2
a.e., and
(6.9)
I'3f (2 From (6.9) we first conclude that
then that the complex dilatation
equals
a.e. (K I )/(K + I). Thus f is afline, hence f = w, and the theorem is proved.
6.5. Distance of Group Isomorphisms from the Identity

Let us represent the torus S = C/G as the point w1/w2 of the upper halfplane, where as before (w1,w2) is a normalized base of G. We recall that if we
= aw1 +
of 0, then
bw2 and
+ da,2 for a, b, c, dE7L with ad be 1. This implies that
is the image of
under the Mbius transformation z h(z) =
change (w1,w2) to another normalized base
(az
+ b)/(cz + d) belonging to the elliptic modular group f.
Let S' C/G' be another torus and 0: G i 0' a given isomorphism.

If 0 is induced by a sense-preserving homeomorphism
S ' S' with a
normalized lift f, the numbers f(w1) and f(2) do not depend on the partic-
ular choice of q (and hence of f). Also, if

and h are as above, it
Wf('2) = h(f(w1 )/f(w2)) for the same h this is seezj from
follows that
f(mco1 + nw2) = n!f(co1) + nf(w2), m, neZ. We conclude that the hyperbolic
distance in the upper half-plane between the points

f(Wk)/,f(w2)
depends solely on the isomorphism 0, not on the base used for 6
the
choice of the generator f. We denote this distance by 5,. It measures how
much 9 deviates from the isomorphisms induced by conformal mappings.
Lemma 6.3. Let 0: G ' 6' be an isomorphism generated by a normalized Kquasiconfornial mapping f. Then
594logK.
Equality holds
(6.10)
and only Vf is the affine transformation generating 0.
PROOF. Let w be the affme nonnalized mapping which generates 0. If w(z) =
p1), we see from (6.5) that

= It' TI/It' where r =
= w(w4)/w(w5). Hence, if K is the maximal dilatation of w,
A.(z +
_____
219
6. Teichmllcr Space of a Torus
logK =log1
=logfr,
I'
20.
=
Consequently, (6.10) holds as an equality for the affine mapping. Inequality
(6.10) and the "only if" part of Lemma 6.3 follow from Theorem 6.3.
0
1
6.6. Representation of the Teichmller Space of a Torus

Let us start again from a fixed torus S = C/G and from a normalized base
In considering
of G. Let p be a point of the Teichmller space
representatives of p we may restrict ourselves to mappings of S which have
a normalized lift f; this follows from Lemma 6.2. We take such a q.' e p and set
= f(wj)/f(w2).
We show that is a well defined mapping of T5. Let us consider another
mapping q' p with the normalized lift f'. Then there is a conformal map
with a normalized lift such that c'o q,' is homotopic to 'p.
c: 4)1(S)
The lifts of 'p and. 004)' induce the same group isomorphism. But the lift
of ci is of the form z , ).z, and so
i = 1, 2. This shows that
f(w1 )/f(w2) does not depend on the choice of 'p p. From the discussion
lies in the upper
after Theorem 6.1 it follows that the image point
half-plane H.
The change of (w1,co2) to another normalized base of G means that iJ' is to
be replaced by ho where h is an element of the elliptic modular group. This
follows from what was said in 6.5 before Lemma 6.3.
-
Thm6.tThemappingl':T5+H,definedby..
=
left of 'p, is a bijective isometry of
where f is the
half-plane fu*Lshr4 with the hyperbolic metric.
(6.11)
onto the upper
PROOF. The mapping

is injective: If fr(p1) =
there are mappings
i = 1,2 whose normalized liftsf, satisfy the equations f1(w,) = Af3(w,),
s(z), then 11 and sof2 determine the same group isomori = 1,2. If z
phism. It follows that their projections q1 and 0 4)2 are homotopic. Here ci,
as the projection of s, is conformal, and so Pi P2.
The mapping is swjective: Given a point z H, we choose an arbitrary
non-zero complex number
After this w set
=
By Theorem 6.1,
there is a homeomorphism 'p of S whose lift f is normalized and has the

properties
=
i = 1,2. From the discussion following Theorem 6.1 it
follows that 9) is sense-preserving. It determines a point p T3, and we see
that
= z.
The mapping is an isometry: Given two points e
i = 1,2, consider
220
their representatives
with the normalized lifts j. Let 4,1(S) +
be
the extremal in the class of quasiconformal mappings homotopic to 4'2 0
and having normalized lifts. For the Teichmller distance is we then have
t5(p1, P2) = log K, where K is the maximal dilatation of On the other
hand, if f is the normalized lift of it follows from Lemma 6.3 that
K
is equal to the hyperbolic distance of the points 11 (o i)/fi(w2) and f(f1 (w1 ))/
f(f1(w2)). But f of1 induces the same group isomorphism as 12' and so
f(f1(w1)) = f2(wj, i = 1,2. Thus the Teichmller distance ts(pi,P2) coincides
with the hyperbolic distance between
and iJ'(P2), and the theorem is
proved.
.0
We know that every equivalence class [4,] has a representative whose lift f
is an affine transformation * +
the (constant) complex dilatation
,z
is of absolute value <I and depends only on
This fact makes it
possible to express (6.11) in an explicit ftm. If we denote the point [4,] by

[z3, we obtain
(6.12)
W2+CLY2Z
The expression (6.12) leads to a simple representation of the Teicjimuller
space of a torus.
Theorem 6,5. The mapping
(6.13)
is a bijective isometry of the Teichmller space T5 onto the hyperbolic unit disc
D.
PRooF. The theorem follows immediatelyfrom the fact that (6.12) is a bijective isometry of onto the hyperbolic upper half-plane.
0
6.7. Complex Structure of the Teichmller Space of Torus

By Theorems 6.4 and 6.5, the TeichmUller space of a torus is a complex
1-manifold.We nOw introduce a complex analytic structure for T5 by means
of the mapping (6.13). The use of (6.11) woWd of course lead to the same
structure, because z
is a conformal mapping.
It follows that the Teichmller spaces of tori art not only IsoMetrically but
also biholomorphically equivalent. We show that a quasiconformal mapping
always induces such an equivalence. Recail that -any
tori are quasiconformally equivalent.
1'
Theorem 6;6. Let S and S' be

and w: S -+ N' a
mapping.
[4,0
Then the bijective isometry [4,] *
of T3 onto T5. is biholomorphic.
221
7. Extremal Mappings f Riemann S9rfoccs
PROOF. The isometry depends only on the homotopy class of w. Therefore, by
Theorem 6.3, there is no loss of generality in assuming that w has a lift

+
Then qow1 has the

We may assume that the lift of is
+
complex dilatation e(z
It follows that
pz), where =
{pow'] induces the contormal self-mapping z
of the unit disc.
['p] +
p)/(l
gz)
By Theorem 6 6, which completes the proof of Theorem 5.5, we can speak

of the complex analytic structure of the abstract TeichmUller space T1 of tori.
It is of particular interest hat under the mapping (6. 3), which defines the
complex .analytic structure for
the Teichmller metric agrees with the
hyperbolic metric of the unit disc.
Teichmller spaces 1,, p> 1, connecbetween the metric and the complex analytic structure will he studied
11) 9.5 and 9.6.
Another interesting result is the 'oncrete

obtain for the modulat group Mod(S) of
1L
from our conside;ations that two points
are
Mod(S) if and only if their images
P2 of
Pt
tji(p1) and
in H are equivalent tinder the efliptic modular group F. In
fact. h
h
is an isouioiphisrn of F unto Mod(S). (Cf., also. the
R5
K/F and
If with Theorem 2.7, which says that
-
'the isomorphisin Mod(Sj
the- discontinuous nature -of
Mod(S). The discontinuity

is true of all compact surfaces S but
more difficult to rove if the genus of S is > 1 (Kravetz [1], Abikoff [2)).
gotfr
I or cs-cry Mohws transfori t n q whtLh maps H onto itself t1s
is a hiholomorphic self-mapping of
It follows- that the modular group is a
proper subgroup of the full group tif biholomorphic automorphisms of the
Teichmller space of a torus. lu 'fr, p> I. the situation is different, as we

shall see in 9.6.

7.1. Dual
Spaces
Having discussed the Teiclim

genus 1, we shnfl no.v
is greater than l. lLcir
r
of a torus, because
no
covering group.
in order to get an insight into

compact surfaces,
shall
Riemann surface of
cu"ipact
whose genus
are iutt as accessible as the space
rcrrescntation for the
a
in
of
in
spaces of
8 the extremal
222
V. TeichinOller Spacee
quasiconformal mappings whose maximal dilatations determine the distance
between. pQints in the 'Teichmller space. This is equivalent to considering

mappings which are extremal forgiven boundary values. Some preliminary
remarks were made in 1.5.7 and in 2.2 and 3.7 Of this chapter. A good part of our considerations can be done in the general seKing
The special case of compact surfaces will not be taken up until the
section 7.8.
Let S be a Riemann surface whose universal covering surface is conformal-
ly.quivaledt to a disc. In this section, we take the universal covering surface

the unit disc D. As before, G denotes the covering group of D over S.
We shall derive a necessary condition satisfied by the complex dilatation of
the extremal mapping. The condition will be in terms of notions related to
Li-spaces. First, we introduce the space
of all bounded measurable
(1,1)-differentials of G. As we have noted before
is a complex
Banach space in which the Beltrami differentials of G form the open unit ball.
Next we consider quadratic differentials of G, i.e., measurable functions
on D which satisfy the condition ((p0 g)g'2 = for g c G. We define the
L1-norm
IIq'II
JN
whr:e N is a Dirichlet region for G. Since Iq' I is a (1, 1)-differential of G (cf.
IV.l.4), this norm is independent of N. The linear space L'(G) of quadratic

differentials of G with a finite L'-norm is also a Banach space.
If E
and (p EL1 (G), the integral of the product
is well defmed on
S (cL IV.l.4). We write
= JN
f jup.
and
coincide with the
and
L1-spaces of N, it follows from standard results on If-spaces that the mapping
is an isomorphism of
onto the dual of L1 (G) (cf. Dunford
Because
Schwartz [1], p. 289).
7.2. Space of Integrable Holomorphic Quadratic

Differentials
Let A(G)denote the subspace of L'(G) whose functions are holomorphic. l1he
set A(G) is closed in L'(G). For if D, = D(z0,r) = {zIIz
<r} is contained
in D and oeA(G). then by the mean value theorem for analytic functions,
(7.1)
It follows, since {g(N)Ig E G} is locally finite in D, that if
and
-P4)
223
7. Extremal Mappings of Riemann
in L1(G), then (choosing
suitably from its equivalence class) q,,
locally
p is holomorphic. We infer that A(G) is a
uniformly in 0. Hence,
Banach space.
Let 1 be a bounded set in A(G). From (7.1) we see that the functions 4E(1)
are locally uniformly bounded in absolute value in D. Consequently, (I) is a
normal family, and so every. sequence of functions p,, e 'b contains a subsequence
which converges locally uniformly in D to a limit function
It follows from Fatou's lemma that
(7.2)
The functions
are saId to converge weakly to iii A(G), i.e., the norms of
are uniformly bounded and urn
4, locally uniformly in 0.
Convergence of a seqenoe in 4(0) implies its weak convergence, whereas
the converse is not always true.
In 4.2 we introduced the Banach space Q(G) consisting of holomorphic
quadratic differentials of G for wbbb the hyperbolic sup-norm
fl(P lie
z.N
is finite. IfS is compact, the spaces A(G) and Q(G) have the same elements: In
this case the closure of N lies in D so that every holomorphic quadratic

differential of G has a finite V-norm and a finite hyperbolic sup-norm.
In certain other cases we can readily compare the spaces A(G) and Q(G).
and integrating over N, we obtain the inequality
Writing I I =
JN
denotes the hyperbolic area of S (cf. IV.3.6). Hence, if this area is

finite, then Q(G) A(G).
On the other band, it follows from (7.1) that
where
ir(1
JD
Hence, if G is trivial, then A(G)
Q(G). The example q(z)

that A(G) is properly contained in Q(G).
(1
z)2 shows
7.3. Poincar Theta Series

There is a classical method for producing quadratic differentials from analy-
tic functions. 1ff is holomorphic in 0, then
a Poincar theta series of f.

Let A(l) denote the space 4(0) in case G is the trivial group, Le., A(1)
consists of functions holomorphic and integrable in 0.
is called
224
V. Teichrnllet Spaces
7.1. The mapping f , )f is a continuous, linear surjection of A(1)

ontoA(G) of norm
PROOF. Let
f6A(l), z0eD, and r = (1

There exist G.equivalent
t..
such
that
D(z0,
2r)
N1
For every
...,
Dirichiet regions N1,

z e D(z0, r) we have
mr2
OEG
JDr(2)
k
Ill
.JD
where (7.!) yields thefirst inequality. It follows that
is locally
uniformly convergent in D.
From this we conclude that OfeA(G). Also,
Jfq
JN
This implies continuity of the mapping f ' f, since its linearity is clear. To
prove c,urjectivity requires more analysis, and we refer to Lchner [2].
and feL' in 0, then
J0pf
Hence, for
Jg(N)
qeG$N
and fEA(l), we arrive at the relatioii
JO
(7.3)
JN
which will have applications later.
7.4. Infinifesimally Trivial Differentials

Let N(G) denote the subset of
which is orthogonal to il(G), i.e., for
whose elements
=0
JN
e ,1(G). Clearly N(G) is a closed linear subspace of

which belong to N(G) are called infinitesimally trivial.
for
Differ-
extend all functions

to the plane by setting
0
We denote by
the quasiconformal mapping of the plane which
complex dilatation and is so normalized that
I ct
ouside 0.
z)
0.
225

is conformal in F = {z I izI>
Then p determines f,, uniquely, and
for
the Schwarzian derivative.
before, we use the notation s,, =
I }.
As
In section 4 we studied extensively the mapping [jij

of the TeichmUller
space into the space Q(G) of quadratic differentials, the different choice for
the universal covering surface and for the normalization of which we have
made here is of course unessential. We called a differential p trivial if it

determines the origin of
Trivial differentials p can also be characterized by
the property that the Schwarzian derivative vanishes identically. Infinite-
simally trivial differentials admit a similar characterization. Note that if

and w is a complex number, then wp is a complex dilatation
<
Theorem 7.2. A dWerential p is infinitesimally trivial If and only If
=o
Jim
W
w-.O
for every z
E.
PRooF. In 1.4.4 and 11.3.2 we established the representation formula
z+
with
Tp(z)=
From this we obtain by straightforward computation
urn
w-.O
because
=
Jr
z)
we are now dealing with points z for which
>
1.
Hence
urn
w-0
where
(n+
1)(n
+ 2)(n
Therefore, the theorem follows if we prove that

0,
0, 1,2,...,
if and only ifpN(G).

If
is the 0-series for f(z) = z", then by formula (7.3),
(7.4)
226
V. TeichmUller
JJD
By Theorem 7.1, the function
then (7.4) follows.
JJK
is an element of A(G). Hence, if peN(G),
Conversely, if (7.4) holds, it follows from an approximation theorem of

A(1). Formula (7.3)
Carleman [1] that p is orthogonal in D to all functions
then shows that p is orthogonal in N to all functions 01 with f in A(1). By
Theorem 7.1, every element of A(G) is of this form, and so jteN(G).
A slight modification of the above proof gives the following result: A

d(fferential p is infinitesimally trivial if and only tf Tp vanishes identically in E.
In section 5 we studied the function p i A(p) = in the unit ball B(G) of
and proved that it is holomorphic. It follows that A has a derivative

DA ' 'i) for every p e B(G) (cf. the definition in 5.1). Direct computation shows
DA(O)(p)
urn
w-0 W
Hence, Theorem 7.2 says that the set of infinitesimally trivial
of G
is the kernel of the mapping DA(O).
In this section we shall use the class N(G) for studying extremal mappings,
but infinitesimally trivial complex dilatations are also met in connection with
other problems in the theory of Teichmilhler spaces. Their importance was
noticed already by Teichmller, and they were utilized by Ahifors and Bets
in their studies regarding the complex structure of Teichmller spaces. For
information about the class N(G) in the Teichmller theory we refer to the
surveys Royden [2], Earle [2], and Kra [2].
7.5. Mappings with Infinitesimally Trivial Dilatations

If the complex dilatation of f9 is in N(G), we can improve the estimate
liSp lIq
611 p
derived in 11.3.3.
Theorem 7.3. Iffy has an infinitesimally trivial complex dilatation, then

(7.5)
PROOF.
Fix a point z e E and consider the holomorphic function

w .
,fr(w) = (lzl2
1)2S111111(z)
has a zero of order at least

which is bounded in
2 at the origin. Application of Schwarz's lemma to
absolute value by 6, yields therefore
in the unit disc. It follows from Theorem 7.2 that
227
7. Extremal Mappings cf Riemann Surfaces
61w12.
Setting w = p
we get back our function
and the estimate (7.5) follows.
Equality can hold in (7.5). To prove this consider the mappingf defined by
I and f(z) z + 2k(fzj 1) +
is
a
quasiconformal
mapping of the plane with p(z) = 0 in
k
f = f,,
E and p(z) = kz/IzI in D. We see that
f(z) = z + k2/z for izi
rr
p(z)fdxdy =
ri r211
n = 0, 1
drdo = 0,
JoJo
It follows that p is in N(G) for the trivial group. Furthermore, pj)
JJb
lim(1z12
= k and
= 6k2.
We shall now apply Theorem 7.3 to prove an auxiliary result about infinitesimally trivial diatations, which will come into use in what follows.
<2. Then,for 0 t 1/4, there isa
Lemma 7.1. Let VE N(G) and

I 2t2.
[tv] such that II
PRooF. By Theorem 7.3,
24t2.
For 0 t 1/4, we have 24:2 <2. Hence by formula (4.6), there is a complex
dilatation a(t) e [tvj for which
110(t)tLo
7.6.
4IIS1j,iEIIq 42t2.
Complex Dilatations of Extrernal Mappings
Fix a point
here fM fixes the points
T5 and consider the family
1, i, 1. In other words, we consider all quasiconformal self-mappings of D
and which agree with an f" on 5D.
whose complex dilatations are in
As has been noted repeatedly, ibis family contains one or more extremal
mappings, i.e., mappings with the smallest maximal dilatation (Theorem 2.1).
The proof of the following lemma uses ideas applied by Krushkal [I] to a
special case and later elaborated by Reich and Strebel [1]; see also Reich [I].
Lemma 7.2. Let
be extremal in its equivalence class, if p
N(G), then
IkL.
PROOF. Set
k, H,cL =
cannot then be extremal.
k1.
We assume that k1 <k and prove
228
Writing v = p
values oft,
K, we
first prove that for all sufficiently small positive
f2flLo(ftv)1
a smaller maximal dilatation than
Lemma 7.1 we shall then
and still keep the maximal dilatation
correct the boundary values of
smaller than that of f
has
Direct calculation gives
= p(z)
1
tv(z)
tp(z)v(z)
Ip(z)I
Re(p(z)V(z))t + 0(t2).
(7.6)
Iiz(z)I
= f"(z), and the remainder term 0(t2) is uniformly bounded in z.

Write E1 = {zDIIp(z)I <(k + k1)/2}, E2 = D\E1. In E1, lji(z)I is strictly
Here
less than k. By (7.6), there are positive numbers 1 and t1, such that for z E E1,
IMC)l
<k1tift<t1.InE2,
1
p1
k2)(k
(1
k1).
From this and (7.6) we deduce that there are positive numbers c52 and t2, such
that for z-E2, fA(4jj < k 52t if t < t2. If 6 = min(61,62), t0 = min(t1,t2),
we thus have
12(C)I<kot
for t .< t0,a'tevery point
The mapping jA need not agree with fL on the boundary. But if 0(t) is as
in Lemma 7.1, then ft = fA of(T(t) has the same boundary values as f $4 We
have
k & + 12t2
+
Ilcr(t)IL,
Hence, for t small enough, II
II
<k =
and
12t2
the lemma is proved. 0
With the aid of this lemma, the desired characterization of extremal comand
(peA(G). We write
plex dilatations can be readily given. Let us assume that
JN
and denote by
= 1)
the norm of p as an element of the dual space of A(G). If G is the trivial group,
this "dual" norm is denoted by
We are now ready to prove the main result about extremal complex dilata-
tions (Hamilton [1], Krushkal [1])..
229
7. Extremal Mappings of Ricinann Surfaces
Theorem 7.4. 1ff1' is extremal in its equivalence class, then

Ii
aH
(7.7)
if
PROOF. By the HabnBanach extension theorem (DunfordSchwartz [1], p.

and lI).II =
63), there is a linear functional in L'(G) with AIA(G)
such that
From what was said in 7.1 we deduce the existence of a
we have to prove that
Since
=
and hAil =
liPlix,
From AJA(G) =
we conclude that
JN
for 4EA(G). Hence
IIKL.
KEN(G), and by Lemma 7.2,
The necessary condition (7.7) for p to be extremal is also sufficient. Reich

and Strebel [2], with the aid of their "main inequality", proved that this is so for the trivial group. Later Strebel ([2], [5)) generalized the result, first to
finitely generated groups and then to all cases.
Theorem 7.4 allows the following conclusion (Kra [2]): Let f1' be extremal
for its boundary values among all quasiconformal mappings. If the operator
not in
0: A(l)+ A(G) is of norm <1, then
By formula
PROOF. Suppose that p e
i}
111111* = sup {[f
< hlpLL0.
f1' is extremal.
This contradicts Theorem 7.4,
In particular, the extremal pin
is not extremal in
0
In terms of
Teichmller distances, the result can be expressed as follows: IfS = DIG, then
[p]) (cf. 3.2 and 3.7).
(ii T5)(O, [p]) <
7.7. Teichmller Mappings

By Theorem 7.4, there
Let f1' be an extremal mapping for the point [p] e
A (G) with if ii = 1, such that
is a sequence of functions
urn I
Jrq
Such a sequence
is called a Hamilton sequence for p.
which converges
A Hamilton sequence always contains a subsequence
weakly in A(G) to a function q, e A(G)(cf. 7:2). It thay happen that q vanishes
is said to be degenerate.
identically, in which case the sequence
230
V. Teichmulier Spaces
If no Hamilton sequence is degenerate, we can draw a remarkable conclusion about the extremal mapping (Strebel [3), Reich [1)).
Theorem 73. Let jt be an extremal complex dilatation for which no Hamilton

sequence is degenerate. If
> 0, then every weakly convergent Hamilton
sequence tends in L1(G) to the same holomorphic quadratic differential
and
(7.8)
PRooF. Let (q,,) be a weakly convergent Hamilton sequence for and

=
urn q,,(z). In 7.2 we showed that E A(G) and II
1 (formula (7.2)). By the
triangle inequality
4'H
(7.9)
Given an c > 0, let F be a compact subset of a Dirichiet region N of G, such
that
<e.
(7.10)
J N\F
Since
q(z) uniformly on F, there is an n0 such that for n> n0,

14'l)) <
4)1 (I4)NI
From
(Iq',,I 14)1)
(7.10) we conclude that
+ lq't
IqI) = 214)1
and from
14)1)) < 2e.
4)1
J N\F
Hence,
SN
for n > n0. From this and (7.9) it follows that

urn
Here 114)11 =
1.
=1
(7.11)
114)11.
In order to prove this, we suppose that 114)11 < 1, and form

4)11). Trivially,
the sequence ((q,,, (p)/
JN
4)n
II,tL.
11
lI4)N4)I \
JN
114)11
231
By
(7.11), the right-hand expression tends to
as n
Hence,
is a degenerate Hamilton sequence for p, which contra
'
in
dicts the hypothesis. It follows that flq'II 1, and
by (7.11),
L1(G).
We conclude that
II
II
= lim
urn
qi)
(p
fl =0.
JN
Consequently,
0=
.JN
(PJ
JN\
This is possible only if the bracketed expression in the right-hand integral is

zero, and (7.8) follows.
Then we
Finally, let be the limit of another Hamilton sequence for
is
Hence, the meromorphic function
deduce from (7.8) that o/ifr
= I it follows that =
a positive constant. From II(PII =
0
A quasiconformal mapping which is conformal or whose complex dilatation is of the form (7.8) is said to be a Teichmller mapping.
In 2.3 we constructed geodesic lines in T5. Suppose that an extremal for

[1J e is a Teichmller mapping with p determined by (7.8). From lormula
(2.4) we see that a geodesic ray from the origin through
to the boundary
of T5 allows the simple representation
0t<l.
In particular, in
ary values of
the complex dilatation

for every t, 0 t < 1.
is
extremal for the bound-
7.8. Extremal Mappings of Compact Surfaces

Theorem 7.5 is still implicit, but there are special cases in which it is possible
to deduce that Hamilton sequences cannot degenerate. In particular, this

conclusion can be drawn if the Riemann
S is compact. We prefer to
express the result in terms of the mappings of S. and recall that studying the
p} is equivalent to studying a homotopy class of quasiconforfamily {
mal mappings of S (cf. 3.1)
I
Theorem 7.6. On a compact Riemann surface, a quasiconformal mapping with

the smallest maximal dilatation in its homotopy class is a Teichmller mapping.
PROOF. Let (q,,) be a weakly convergent Hamilton sequence with limit (p. For
a compact Riemann surface of genus > I, the Dirichlet regions N are relatively compact in B (IV.5.1). Therefore
232
V. leichmller Spaces
l=limfI4,nl=
,JN
is not degenerate and the result we wanted to prove follows from

Flence,
Theorem 7.5.
0
Since we assumed in this section that the Riemann surface S has a disc as
its universal covering surface, our reasoning does not apply as such to a
torus. But this case was thoroughly handled in section 6. It follows from
Theorem 6.3 that also in the case of a torus the extremals are Teichmller
mappings.
Theorem 7.6 is Teichmller's existence theorem. The first proof,, which is
[1]
based on a continuity argument, is in Teichmiiller [2]; see also
and Bers [3]. Apart from technical details, the reasoning we used to prove
Theorem 7.6 is due to 1-lamilton [I].
Extremal mappings of compact Riemaun surfaces will be studied more
closely in the following section, where Theorem 7.6 will be complemented by
another famous result of Teichmller: On a compact Riemann surface, a

Teichmller mapping is always the unique extremal in its homotopy class.
Our proof of this uniqueness theorem will be based on the original reasoning of Teichmller [1], with due regard to the clarifications later made in it
by Bers [3]. It would also be possible to apply the reasoning which yields the
converse of Theorem 7.4. Strebel [5] showed that such a method also establishes the uniqueness of the extremal.
8. Uniqueness of Extremal Mappings of

Compact Surfaces
8.1. Teichmller Mappings and Quadratic Differentials
Having classified in !V.4. 1 all Riemann surfaces which admit the plane as a
universal covering surface, we know that if a compact Riemann
is not
the sphere or a torus, i.e., if its genus is > 1, then it has a disc as it's universal
covering surface. The covering group is finitely generated, and its Dirichiet
regions are relatively compact on the universal covering surface (IV.5.1).
In section 4 we proved that
' S,IH is an imbedding of the Teichmller
space T5 into the space of holomorphic quadratic differentials. Let us now
assume that S is a compact Riemann surface with genus p> 1. Then there is
another way to associate a holomorphic quadratic differential with each

point of T5. This leads to an imbedding which is simpler than the one given
by the mapping [p] -+
in that the image of T5 is $he open unit ball.
233
8. Uniqueness of Extremal Mappings of Compact Surfaces
In order to associate quadratic differentials with points

space
the Teichmller
of a compact surface, we return to the extremal problem treated
in section 7. We proved that in every homotopy class of sense-preserving

homeomorphisms between two compact Riemann surfaces, a mapping with
the smallest maximal dilatation is always a TeichmUller mapping (Theorem
7.6). More precisely, an extremal mapping f of a surface S is either conformal
or its complex dilatation is of the form
(8.1)
0 < k < I and q is a holornorphic quadratic differential of S.

In this section we shall prove that every Teichmller mapping is the
where
unique extremal in its homotopy class, i.e., that each homotopy class contains
exactly one Teichmller mapping. This makes it possible to define an injective
mapping of the Teicbmller space into the space of quadratic differentials.
Given a holomorphic quadratic cbfferential q on S. we say that the pair

the Teichmlfer mapping f which has the complex dila-
(q,, k) determines
tation (8.1). 1Ff is determined by another pair 1,k1), we clearly have k1 = k.

isa positive constant. We see
and it follows that
Thus
=
that a non-conformal Teiehmller mapping determines the associated quadratic differential up to a positive multiplicative constant.
We note that the absolute value of the complex dilatation of a Teichmller
k) has the
mapping is constant and that a mapping determined by the pair
maximal dilatation K = (1 + k)/(1 k). The complex dilatation (8.1) is defined
at every point of S. except for the finitely many zeros of
8.2.
Local Representation of Teichmller Mappings
In the case of a torus, the properties of extremal mappings can be determined

without great difficulty. This is due to the fact that on a torus, all holomorphic
quadratic differentials are constants. The induced metric is therefore euclidean,
and the universal covering surface, the complex plane, with its explicit
covering group offers a convenient framework for studying extremal mappings. We proved that on C the extremals are globally affine, and that, up to
translations, they are unique in any homotopy class.
If the Riemann surface S has genus p> 1, the situation is much more
complicated. Holomorphic quadratic differentials now have 4p 4 zeros,
which means that the induced mctric has singularities. Global coordinates
cannot be used for the study of Tcichmller mappings. However, there is a
certain analogy with the case of a torus. It turns out that in a sense which we
shall now make precise, every Teichmller mapping is locally affinc.
The local behavior of a Teichmllcr mapping f becomes clear when we
introduce a suitable quadratic differential on the surface S' onto which J
maps S (Tcichmller [1]).
V. Teiehrnllcr Spaces
234
S' be a Teiehmller mapping determined by a pair

on S'
(q, k). Then there exiSts a unique holomorphie quadratic d.fferensial
properfies 1 if p has a zero of order n 0 at p. then #
with the
2 If is a natural parameter of at a
the same order a:
has a zero
regular point p, then the mapping f has the representation
Theorem 8.1. Let f: S
(8.2)
in a neighborhood of p in terms of a natural parameter
of
at f(p).
PROOF. Let p be an arbitrary point of S. (We actually prove a little more than
what is stated in the theorem.) Assume that q, has a zero of order n at p,n 0
0), and denote by a natural parameter
(n = 0 means of course that
of at p. We define a mapping in a neighborhood of f(p) by setting
of
2/(11+2)
+
=
Ik
- (8.3)
If p is a regular point, (8.3) reduces to the simple form (8.2). We prove that C'
is a local parameter on the Ri.mann surface S'.
Let z' be an arbitrary local parameter on S' in a neighborhood of the
Since p(z) dz2
point f(p). Then z' of has the complex dilatation
((n + 2)/2)2 111dC2 (formula (6.1) in !V.6) and sncc complex dilatation is a
( 1, 1)-differential, we see that the complex dilatation of z' of can also be
-
expressed in the form k(C/JC9".
On the other hand, we compute from (8.3) that the mapping of has the
complex dilatation
It follows that
is
a
local
parameter
of S'.
C'
In order to construct the differential st', we consider a point
p which
is so close to p that p2 is a regular point and lies in the domain of C. A natural
dz = ((ii +
parameter at Pi is obtained if we integrate
dC It
follows that =
is a natural parameter at Pi. Then
+
=
(I k) is a local parameter of S' at
and
= (n +
We
conclude that
= ((ii + 2)/2)2C'dC'2
defines a holomorphic quadratic differential (i on S'. We see that

natural parameter of at 1(p). It has a zero of order n at 1(p).
If C =
+ Lip,
of =
is
+ iq', the mapping (8.2) assumes the form

(8.4)
with K (1 + k)/(I k). Hence, formula (8.2) can be expressed as foftows

a neighborhood i1 a regular point of the associated Thchmi,lIer mapping)s
235
8. Uniqueness o(Extrem& Mappings of Compea Surfaces
a conformal transformation, followed by a fixed stretching in the direction of

the positive real axis, followed by another conformal mapping.
For the Teichmller mapping f: S -. S'. we call rp its initial and its
terminal differential.
8.3. Stretching Function and the Jacobian

The stretching (8.4) is a characteristic property of a Teichmller mapping.
We shall now define a stretching function for an arbitrary quasiconformal
mapping of S.
Let p be a holomorphic quadratic differential on the Riemann surface S.
On S we use the metric induced by p (cf. IV.7 where such a metric was
studied). In addition to S. we shall consider in the following the image S' of S
under a Teichmller mapping with q as initial differential. On S' we use the
metric induced by the terminal quadratic differential of this Teichmller
mapping.
Let f: S -, S' be a quasiconformal mapping and z ' = w(z) its representation in a neighborhood of a regular point p ES. We wish to dime a stretching function of f on S. so that if z = x + iy and z' are natural parameters,
we have
This is achieved if we set

21(p)
8w(z)
Ow(z)
k('(w(z))l
(8.5)
first note that since and are quadratic differentials,

is a Borel
function on S. If z and z' are natural parameters, q =
=
1, and so
We
low + Owl =
Let a be a horizontal arc on S. The differential Aj.lcol"2 can be integrated
along a (with the possible exception of a family of arcs a whose union has the
area zero), and from the representation A1 = 1w,] it follows that
= l(f(a)).
(8.6)
In later integrations we also need the Jacobian with respect to the o- and
It is defined by the formula
J1(p)
(lOw(z)J2
We see that
is a function on S. In natural parameters,
ordinary Jacobian 10w12 10w12. Therefore,
f
where ISI denotes the
area of S'.
becomes the
V. TcichmHer
236
8.4.
Average Stretching
The following result about average stretching is an essential step on the road
leading to the extremal propcrties of Teichmller mappings.
Lemma 8.1. Let f: S . S be a quasiconformal mappiny hornotopic w the identitv and q a holomorphic quadratic
on S. Define A',. by (8.5)for ahd
= q,. Then
Js
a one-dimensional average of A = 2,. Let

of a horizontal trajectory with midpoint p and of length 2a. We set
PROOF.. We first define
be a subarc
(8.7)
2aja
Assume, for a moment, that q has an oriented trajectory structure (cf.

be the union of the non-critical horizontal trajectories of q).
Since there are only finitely many critical horizontal trajectories on S, the set
S\SO has the area zero.
We define a flow on S0. tet p be a point on a horizontal trajectory
and i a real number. Let
t) denote the point on which has the distance
lV.6.3). Let
It) from p and lies in the positive direction from p if t > 0, in the negative
direction if <0. Then x: S0 x R . S0 is a continuous mapping such that
x(p. t) is a conformal and isometric bijection for every t.
We conclude that A! = A o x(. t) is a Borel-measurable function on S0, and
1 1114,1
Js
Js0
2114,1
M4'I =
Js
l-fene,
where Fubini's theorem justifies the last step. Here
=
We
have thus proved that
Js
provided that
q,
(8.8)
Js
has an orientable trajectory system.
If the trajectory system of 4, is not orientable, we construct as in IV.6.3 a

two-sheeted covering of S on which the trajectories of the lift of (these
of (
of Extremal
8.
ate the
of;. and
the
on
Surfaces
and denote the lifts

can he oriented.
then it is dear from ftc construction of that
of
J&
Js
It follows that (8.8) is always
From (87).
(8.6),
and Lemma
2a(P)
that
we
M/a
almost everywhere, where the number M does not depend on
(1
Hence, by
(8.9)
M,"a)jSI.
,Js
On every non-critical trajectory,
rviaUer whether it is a spiral or a closed
curve, wc can take a as large as we please. Letting a
lemma from (8.9).
x. we obtain the
8.5. Teichmile(s Uniqueness Theorem

We have now completed the preparations needed for our proof of Teichmuller's basic result about the uniqueness of extremal quasiconformal
mappings of compact Riemann surfaces.
Theorena 8.2. Let S he a compac: Riemann surface of genus > 1. In every
honiotopy class of sease-preserv(ng homeomorphisnis of S onto another Riemann
surface S', there is exact/v one Teich,nller mapping, and its maximal dilatation
is unique(y smallest.
By Theorem 15. every homotopy class of sense-preserving homeomorphisms of S onto 5 contains quasiconformal mappings. By Theorem 2.1, each
contains a mapping with the smallest maximal dilatation. By Theorem
7.6, every such extremal is a Teichmller mapping. Consequently, it remains
to be proved that every Teichrnller mapping is the unique extremal in its

homotopy class. Taken together the results then imply that each homotopy
class contains exactly one Teichmller mapping.
is also
Suppose first that
S
S -.+
conformal and homotopic to
then f 'of0: S S is conformal and homotopic to the identity. By Theorem l.3,f =
and so f0 is a unique extremal.
Henceforth we assume that f0: S + S' is a Teichmller mapping defined by
the pair (o, k0), k0 > 0. Let f: S p S' be a quasiconformal mapping homoto-
pic to
II K0 and 9K denote the maximal dilatations of
first that
and f, we prove
V. Tcichm*fler Spaces
KK0.
(8.10)
The quasiconformal mapping h =
8' + S'
is homotopic to the
identity. We apply Lemma 8.1 to h, using on S1 the metric induced by the

terminal quadratic differential of 10. It follows, by Schwarz's inequality,
that
(8.11)
In order to estimate the left-hand integral we consider a regular point p of

S with the additional property that f(p) is a regular point of S' for These
conditions exclude only finitely many points p of S. Let C = +
be a
natural parameter at p. and C0 and natural parameters at the points

and f(p), respectively, such that in terms of C and C0, the mapping f0 has the
representation
Co =
1_no
k0 =
+ in,
K01
(8.12)
We see that
and
are both equal to the constant K0.
Let
= w(C0) be a representation of h. (The various mappings are illustrated in Fig. 15.) By (8.12), f has the representation
1
Differentiation
yields
0C1(C) +
f = hof0, we conclude that
Ko(Dw(C0)
Because
h=
S.
Figure 15
39
8. Uniqueness of Extremal Mappings of Compact Surfaces
=
almost everywhere.
Reverting to the integral in (8.1
we now obtain
=$
Next we take into account the fact that f is K-quasiconformal. We express

the dilatation condition in the natural parameters and and deduce that
A KJf
(8.13)
a.e.
Thus
(K/KO)j'
= (K/K0)IS'I.
J114'I
In conjunction with (8.11), this yields (8.10).

Assume now that equality holds in (8.10). The above reasoning shows that
(8.13) must then hold as an equality a.e., so that
= K0.!,-
a.e.
(8.14)
We show that (8.14) is possible only if f and f0 satisfy the same Beltrami
equation.
We use again the natural parameters
and
From (8.12) we first see
that
=
In terms
equation (8.14) assumes the form
+
On the other hand,
is K0-quasiconformal, so that
-.
+
These
relations imply that
Further,
It
follows that
a.e.
K0(IaC1I
a.e.
and
whence
hence
also satisfies the equation
=
We
conclude that the mapping
is conformal, which means that
is conformal. Since
is homotopic to the identity, Theorem 1.3
tells that
is the identity mapping. The unique extremality of 10 has
thus been proved.
0
we already said at the end of section 7, the above proof of Theorem 8.2
follows essentially the modification Bers ([3] and the appendix in [12]) gave
of the original proof of Teichmller [1].
As
V. Teichmuller Spaces
240
Teichmller Spaces of Compact Surfaces
9.
9.1. TeichmUller Imbedding

Let S be a compact Riemann surface of genus >1. Theorem 8.2 provides a
new way to map the Teichmller space of S into the space Qs of holomorphic
quadratic differentials of S. By Theorem 8.2 every point p E T5 contains

Teichmller mappings and they all have, the same complex dilatation. If
p = 0, these mappings are conformal. For all other points, the mappings are
determined by a pair (o,k), where 0 < k < I and oeQ5 is unique up to a
positive multiplicative constant.
By Theorem JV.5.5, the linear space Q5 is finite dimensional. By a fundaare topomental theorem of linear algebra, all norm-induced metrics on
logically equivalent. Of the many natural metrics available, we choose here
the one which is induced by the hyperbolic sup norm: lIq'fl = sup I'pI/'i2.
k), we assume henceIn representing Teichmller mappings by pairs
= 1. For the point peT3,
forth that p is normalized by the requirement
p 0, we then have the unique representation
p=
this case k =
0,
and q is not
uniquely defined, but it is of course immaterial how q, is chosen.

(9.1)
where (peQs,
unit ball of Qs
= 1, andO k < 1, is a homeomorphism ofT5 onto the open
PROOF. By our previous remarks, (9.1) is a well defined injection ofT5 into the
open unit ball of Qs. It is clearly also a surjection onto this ball.
In order to prove that (9.1) is continuous, we assume that

in
We use in its fl-metric (see 111.2.2), which
p=
of T5. By Theorem
is topologically equivalent to the Teichmller metric
k. Hence, by the triangle inequality
8.2, we have PS(PR,O) = k1,,
It follows that
k = 0, this implies that

ki
koll
Assuming that k> 0 we show that pM *
in Qs. Let us' lift the differentials
and qi to the universal covering surface H. In H, the analytic functions

n = 1, 2, ..., constitute a normal family. If
does not tend to
o as n p
thre is a subsequence
and a differential e Q5, t/i
a.e. By Lemma
0. Then
fl#II = 1, such that

3.1,
Hence
,fr,
which
is
a
contradiction,
and so
q
=
241
9. Teicbmiler Spaces of Compact Surfaces
urn
0.
From
. k(P, i.e., that (9.1) is continuous.

we finally see that
The continuity of the inverse of (9.1) follows from the invariance of domain.
The space Qs is finite dimensional and T5 is homeomorphic with an open

subset of Q5 (Theorems 4.3 and 4.8). Hence (9.1), as a continuous injection of
T5 into Q5, is a homeomorphism of onto its image.
0
We call (9.1) the Teichmller imbedding.
9.2. Teichmller Space as a Ball of the Euclidean Space

In proving Theorem 9.1 w chose the hyperbolic sup norm in Q5 for technical
reasons only. Since Qs is a (3p 3).dimensional linear space over the complex
in Qs Then an arbitrary
numbers, we can fix a base 4's,
has a representation
with complex coefficients; = x + iy1. The quadratic differential q can thus

or with the point
be identified with the point (z1,z2
of
of the euclidean space
We now introduce
the eudlidean norm
\1/2
/3p3
H11
+ y?)
in Q5.
Using this new norm, we can rephrase Theorem 9.1 as follows:

(9.2)
= 1,andO k < 1,isahomeomorphism of T5 onto the open

where coQs,
unit ball of the euclidean space
The validity of this statement follows immediately from Theorem 9.1, by

virtue of the fact that all norms in Qs define the same topology.
We conclude, in particular, that the Tcichmller space of a compact Riemann szirfac& of genus p> 1 is homeomorphit to
This is an old
result of Fricke who proved it without the
of quasiconformal mappings
by suitably parametrizing covering groups over compact surfaces. A modernized version of Fricke's proof has been given by Keen [13.
V. TeichmUller Spaces
242
lip 1, we proved that the Teichmller space is isomorphic to the hyperis hobolic upper half-plane (Theorem 6.4). Hence, for p = I, the space
meomorphic to R2
The Bers imbedding of Theorem 4.8 also allows the conclusion that for
However, not
p> 1, the SpacCTS is homeomorphic to a subdomain of
much is known about this domain. Now we can immediately draw the following conclusion: The Teichmller space of a compact Ricman surface is
some of its interest in light
contractible. This special result has, of course,
of the DouadyEarle result that every Teichmller space is contractible
(see 3.6).
9.3. Straight Lines in Teichmller Space

Let S be a compact Riemann surface of genus p> 1. Then every point
of the Teichmller space of S can be represented by the complex dilatation of a unique TeichmHer mapping. If q =
then r3(O,q) =
+ k)/(I k)).
In 2.3 we studied geodesic lines in Teichmller space. Since extremal

mappings are unique, the remark made in 7.7 can be complemented. We
deduce from formula (2.4), by using Theorem 8.2, that
I<zt.czl,
is a geodesic line in T5 which passes through the origin and the point q.
Here
Consequently, by changing the parameter, we obtain the mapping.

x
which is an isometry of the real axis into

line. It follows that every point of
We call such a path in T5 a

lies on a straight line through
the origin.
In order to study geodesics through two arbitrary points of T5, we recall
that all Teichmller spaces of Riemann surfaces of genus p are isomorphic
(see 5.6). The space T5 is a model of the abstract space 1;.
Let q1 and q2 be two points of T5. We can map T5 isometrically onto

Teichmller space such that the image of one of the given points lies
at the origin. In view of what we proved about straight lines through the
origin, we obtain the following result: In the Tekhmller space 1, any two
points lie o' a straight line.
This result holds in the case p =
the hyperbolic upper half-plane.
also, because i; is then isomorphic to
243
9. Teichmller Spaces of Compact Surfaces
9.4. Composition of Teichmller Mappings

For an application in 9.5 we need a result about the composition of Teichmuller mappings. As before, S is a compact Ricmann surface of genus >1.
Let f: S +5' be a Teichmller mapping determined by a pair (q,, k). Then
5' - S is a Teichmller mapping detennined by the pair ( k), where
is the terminal differential off. For it follows from formula (8.2) that near
a regular point of ii',
and are the natural parameters associated with

this we conclude that the complex dilatation off-1 is
where
and ,,1'. From
by making
use of the fact that dC'2 = i/i(z') dz'2 and that the complex dilatation is a
(1, 1)-differential.
Let us multiply (9.3) by iK = i(1 + k)/(l k). Comparison with (8.2) then
yields the result that the terminal differential off 1 is
For composed mappings we prove what will be needed later.
Lemma 9.1. Let f1: S S1 and 12: S + S2 be Teichmller mappings determined
k1) and (q'z, k2) such that q2fq1 is a constant. Then f2 off' is a Teichby
muller. mapping. Up to constants, the initial differential of f2 ofr' agrees with
that offr', and the terminal dfferential with that 0112.
Let pS be a regular point for Pi and

We denote by C and C
natural parameters for (pj and
at p vanishing at p, and by and C2 the
corresponding natural parameters for the terminal differentials of f1 and f2
PROOF.
at f1(p) and, f2(p).
Let us consider the projection mapping ,

composed into mappings
and
C
C
It can be de-
of
and
is constant, we see that C*/C is constant. It
follows that C1
a
C2 is an afline transformation. Explicitly, we write
> 0,0 < 0 < 2ir. Then we can choose
The stretchings
It follows that
--+
and C* ' C2 can be determined from (9.3) and (8.2).
C2
2e
Since C1 is the natural parameter of the terminal differential of f1, we see

from (9.4) that 12 ofr' is a Teichmller mapping whose initial differential is
a constant times the initial differential of fr'. The result about the terminal
differential of ofr' is obtained if we interchange the roles of f1 and 12.
V. Teichmuller Spaces
SI
'I
Figure 16
For the complex dilatation of
,
i_
Let
10
we obtain the expression

7
95
K1 and K2 be the maximal dilatations of f1 and f2. As regards the
we use (9.5) to make precise what seems clear

maximal dilatation of f2
geometrically. Suppose first that the initial differential 42 of is a positive
constant times the terminal differential
i.e., that 0 = x. We
of
then conclude from.(9.5), by simple computation, that 12 ofr' has the maxiis a negative
mal dilatation K1 K2. The other extreme occurs if
constant, i.e., if 6 = 0. It then follows from (9.5) that 12
has the maximal
dilatation max(K1/K2, K2/K1).
The result of Lemma 9.1 can also be expressed as follows: 1ff1: S S1 and
f2: S1 , S. are Teichmller mappings, and the ratio of the terminal differential
of and the initial differential of 12 is constant, then 12
is a Teichmller
mapping.
9.5. Teichmller Discs

In 6.6 and 6.7 we proved that there exists a holomorphic isometry between
the Teichmller space of a torus and the hyperbolic unit disc. We shall now
show that the same is true in a local sense in every Teichmller space 1.
Let S be a coffipact Riemann surface whose genus is 1. We fix a bobpositive conmorphic quadratic differential q of S (up to a
stant) and consider the
(9.6)
9. Ieichmllcr Spaees of
245
of the unit disc D into the Tcichrnller space T5. By Theorem 8.2. this
of D into the set B(G) of

mapping is ir.jective. The mapping
Beltrami differentials of S is holomorphic. As a composition of
(9.6) is Jzolomorphic.
and the canonical projection it. i/ic
Let
=
the image of D under (9.6). The subset A of
disc determined by -p.
denote
is called the Teich,nller
With varying p. cvery A clearly contains the origin of T5. But otherwise
two TeichmUlkr discs arc disjoint if they are determined by quadratic differentials whose quotient is not constant. This follows from the definition of
A, in view of the uniqueness result of Theorem 8.2.
Theorem 9.3. The mapping '
is an isometrv of the hyperbolic unit
disc onto the Teichrnller disc determined by q.
PROOF. For dstances from the origin, isometry is clear:
=
where h denotes the hyperbolic distance in the unit disc.
In the general case, let

and f, be the Teichrnller mappings with complex
and
By Lemma 9.1, the composition 1'
is also a Teichmller mapping. From (9.5) (or from formula (1.4.4)) we see
that its complex dilatation has the absolute value ((zr
Therefore, by Theorem 8.2,
--
I
2
9.6.
+
II
Ji z1z21
z2j
=h(;1,:2).
Z21
Complex Structure and Teichmller Metric
The results concerning Teichmller discs admit far-reaching generalizations.

In order to explain this, we first generalize the notion of hyperbolic distance.
Let f be a holomorphtc function in the unit disc D with values lying in D.
By using suitable Mbius transformations, we see that Schwarz's lemma can
be expressed in the form
h(f(z1),f(z2)) < h(z1.z2).
(9.7)
From the way the hyperbolic metric was introduced to other domains by use
of conformal
we conclude that (9.7) holds if f is an analytic
mapping of I? into an arbitrary simply connected domain with more than one
246
V. Teivhmller Spaces
boundary point, and also, if f maps D holomorphically into a Riemann

surllFce which admits D as its universal covering surface.
Property (9.7) gives rise to one more generalization. Let X be a complex

analytic manifold. There exists a largest pseudometric d on X such that
d(f(z1),f(z2))
h(z1,z2)
all holomorphic mappings 1: D ' X and for all z1, z2 in D. If d is a metric,

it is called the hyperbolic (or Kobayasbi) metric of X (Kobayashi [1]).
The following result of Royden [1] connects the Teichmller metric with
the complex analytic structure of 7,.
for
Theorem 9.4. In the space

metric
p.
I, the hyperbolic metric and the Teichmller
are the same.
1, the theorem follows from (9.7) and the fact that 7, can then
be identified with the hyperbolic unit disc (see 6.7).
PROOF. For p
For p> 1, let us consider a model T5 of

In studying the
between two points q1 and q2 of T5, we may assume, by replacing
by
another modelof 7,, that q1 lies at the origin. Let q2 = [kc/(p1].
Let us apply again the holomorphic mapping (9.6) of the unit disc into i's.
By Theorem 9.3,
= h(z1,z2),
(9.8)
where z1 and z2 are the preimages of q1 and q2 under (9.6). Here the bobmorphic mapping of D into T5 depends on the given points. Therefore, we can
only infer from (9.8) that the Teichmller metric is greater than or equal to
the hyperbolic metric of
Equality is in fact more difficult to establish; for the proof we refer to

Theorem 9.4 says that the Teichmller metric can be recovered from the
complex analytic structure of 7. It follows that the Teichmller metric is
invariant under biholomorphic self-mappings of 1;.
Royden (1] also proved that in 7,, p> 1, the modular group is the full
group of biholomorphic automorphisms of T,.
We showed in 2.7 that two points [ft] and [f2J ofT5 are equivalent under
the modular group if and only if the Riemann surfaces ft(S) and f2(S) are
conformally equivalent (Theorem 2.7). Hence, if S is a compact Riemann
surface of genus greater than 1, then by Royden's result, the points [f1J and
[f2] of T5 are equivalent under bihobomorphic mappings of T5 if and only if

the surfaces f1(S), and f2(S) are conformably equivalent.
Royden's result also shows that the Teicbmlter imbedding (9.2) is not
hobouiorphic. For if it were, it would be biholomorphic. The group G of
would then be isomorbiholomorphic self-mappings of the unit ball of

phic to the modular group of 7;. However, the modular group is knownrto
9. Teichmllcr Spaces of Compact Surfaces
247
be properly discontinuous, whereas G (which contains all unitary mappings)

is not.
9.7. Surfaces of Finite Type

A Riemann surface S0 is of finite type ifS0 is contained in a compact Riemann
surface S and S\S0 consists of finitely many points and of finitely many
disjoint closed parametric discs of S.
Let S be of genus p, and consider the case in which S\S0 is the set of n
points. We say that S0 is then of type (p. n). The points of S\SO are called
punctures.
It follows from our previous analysis of covering groups that a Riemann
surface of type (p. n) admits a disc as its covering surface if and only if
2p 2 + n > 0. Under this condition, which leaves otjt only the tori and the
extended plane with no more than two punctures, the Teichmuller theory of
of compact surfaces
n) is largely analogous with the
surfaces of type
of genus > 1. In particular, the following basic result is true:
Theorem 9.5. The Teichmller space of a Riemann surface of type (p, n),
2 + n > 0, is homeomorphic to the euclidean space
The proof can be reduced to the case of a compact surface. The idea is to
construct suitable coverings of S branched at the points of S\S0. The procedure is explained in Ahlfors [1], pp. 2023.
We have emphasized many times the fact that the Teichmller spaces of
compact surfaces with the same genus are all isomorphic. Similarly, the
Teichmller spaces of all Riemann surfaces of type (p, n) are isomorphic,
because such surfaces are all quasiconformally equivalent.
Finally, let S0 be a Riemann surface of finite type such that S\S0 contains
discs. A Teichmller theory analogous to the theory of compact surfaces can
again be developed for such surfaces. However, in this case the notion of,
reduced Teichmller space (see 2.1) must be used. For an exposition of the"
theory of Teichmller spaces of Riemann surfaces of finite type we refer to
Abikoff [2].
Bibliography
Abikoff, W. 11] Some remarks on Kleinian groups. Advances in the Theory of Riemann Surfaces. Annals of Math. Studies 6&Pnnceton Univ.
IS.
12] The Real Analytic Theory of Teichmller Space. Lecture Notes in Math.
820. Springer-Verlag (1980).
[3] A geometric property of Bers'
of the Tcichmller space. Riemann Surfaces and
Topics. Annals of Math. Studies 97. Princeton Univ.
Press (1981),
Ahlfors, L. V. (liOn quasicon formal mappings. Journal d'Analyse Math. 3(1953/54),
158.
analytic structure of the. space of closed

(21 The
Analytic Functions. Princeton Unjv. Press (1960), 45-66.
surfaces.
[31 Some remarks on Teichmller's space of Riemann surfaces. Annals of

Math. (2)74(1961), 471191.
reflections. Acta Math. 109(1963), 291301.
on Quasiconformal Mappings. Van Nostrand (1966).
151
Abhors,
and Bets, L. [1] kiemann's mapping theorem for variable metrics.
Annals of Math. (2) 72(1960),

-404.
Ahifors, L. V. and Sano, L. El) Riemann Surfaces. Princeton Univ. Press
Ahifors, L. V. and Well), G. [1) A Uniqueness thedrem for Bltrami
Proc.
Amer. Math. Soc. 13(1962), 975978.
Astala, K. and Gehring, F. W. (11 Injectivity, the BMO norm and the universal
Teichmller space. To appear in Journal d'Analyse Math.
1. [1) On a theorem of Mon and the definition of quasiconformality. irony.
Bets,
Amer. Math. Soc. 84(1957), 7884.
(2) Spaces of Riemann surfaces. Proc. International Congress of Matheinaziclans, Edinburgh 1958. Cambridge Univ. Press (1960). 349361.
mappings and
(3]
theorem. Analytic Functions.
Princeton Univ. Press (1960), 89119.
141 Simultaneous uniformization. Bull. Amer. Math. Soc. 66(1960). 9497.

151 Umformization by Beltrami equations. Comm. Pure App!. Math. 14 (1961),
215228.
-(61 Correction to "Spaces of Riemann surfaces as bounded domains". Bull.

Amer. Math. Soc. 67
465-466.
249
(7] On moduh of
Ifloitk. Ek
ituisq.sinS!i!ut
surfaces. Let-ture.s
Mathe
lechnisuhe l-lochsctiule. Zurich 1964).

Complex
Tcichmuller spaces. Proc.
(81 Automorphic forms and
-tneapolis 1964. Sprineer-Verlag (1965). 109- 113.
with applicattons. to quas;conformal
191 A non-standard inteeral
113134.
mappings ic-ta ifat/:. 116
Kleinian groups. Bull. London Math. Soc. 4
moduli.
1101
(1972).257-300.
mappings. with applications to differential equations,
[11]
function theory and topology. Bull. inter. Mat/i. Soc. 83 ij977). 10S3-- 1100.
[12] A
of a fundamental inequality for
Journal d Analtsc
36 (1 979'L I
30.
A I Mat!: 10 (1985).
(13) On a theorem of Abiko!i. . Inn
ad. Sei.
83-87.
A. and Ailfors. 1.. V. [ I

hutmdai
conformal mappir.es. Acza Math 96 ( 195t). 125-flau; bak'L N.
Jn.thulque.s. I
II
XXXI)!. Fascicuk de
is. 1).
quasi
I)
agraphcs
radius of
of
a 7.
i-f
circuiar triangles and regular
lariahles
.4ppl. 4 (1985).
Funktioncn durch lineare
I1bt'r die
o\imation
1,. no. 9 (1923).
Aggregate von vorgegcbcaen Poteuzcn. Ark. Mat.
Complex
T. [1)
A. and
C. J 11] ('onlotinally natur.cl
of the circle. To appear in Acri: Mail,.
l)unford. N. and Schwartz. J. T. (I) Linear
Part 1.
I
(1958).
L)u,cn. P. 1. 11) L'niraknt Functions.
honieomoIphlsn1S
Publishers
1983).
Earle. C. J. [11 The TeichmUller space ot an arbitrary Fuchsian group. J3&1. .4mer.
.4fath. Soc. 70 (1964), 699-701.
[2] Teichmltcr theory. Di.srreie Groups and 1 liwnwrp/zic Functions, ed. W. J.
Harvey. Academic Press(1977). 143.162.

Earle. C. i. and Eells, J. Jr. [I] On the differential geometry of TeichmUller spaces.
Journal d'Analvse Math. 19 (1967). 3552.
Fknn. B. (Rrown) Ill Jordan domains and the universal Teichrnller space. Trans.
Amer. Math. Soc. 282(1984). 603 610.
lord. 1.. R.
Functions. 2nd ed. Chelsea. New York (l951).
Fricke. R. and Klein. F. (I] i.'orlesungen her die Theorre
auwmarpiwn Functionen.
Teuhner (1926).
Gardiner. I-. P. (l)An analysis of the group operation in universal Teichmlier space.
Trapis. .4'ner. Math. Soc. 132 (1968). 471-
Gauss, C. F. [I) Ailgemeine Auflosung der Aufgabe die Thcile einer gegebenen
l'lche auf einer andern gegebenen Flache
abzuhilden. dass die Abbildung
dem Ahgebildeten in den kleinsten 'Fheilen hnlich wird
tronomische ,4/ihand-
lungen. vol. 3 (ed. H. C'. Schumadier). Aliona (1825). or carl Friedrich Gauss
Werke. vol. IV. Gottingen (1880). 189-216.
(iehring. F. W. II)
mappings which hold the real axis ruintwise
fixed. Math. Essays Dedicated to A. J. Macintyre. Ohio Univ. Press. Athens
(1970). 145-148.
(2) Univalent functions and the Schwarzian derivative. Comment. Math. Heft.
52(1977).
561
572.
[3) Spirals and the universal Teichmller space.
Math. 141 (1978),

99-113.
[4) Characteristic properties of quasidisks. .Stn. Math Sup. 84. Presses Univ.
Montral (1982).
Bibliography
250
Gehring, F. W. and Osgood, B. G. [U Uniform domains and the

metric. Journal d'Analyse Math. 36(1979), 5074.
the Nehari univalencecriterlon and
Gehring, F. W. and Pommerenke, Ch. [1]
quasicircies. Comment. Math. Helv. 59(1984), 226242.
Gehnng, F. W. and Visl, J. [1] Hausdorif dimension and quasiconforinal map-.

plugs. J. London Math. Soc. (2)6(1973), 504512.
Grtzsch, H. [1] Uber die Verzerrung bei schhchten nichtkonformen
und ber eine damit zusammenhangende Erweiterung des Picardschen Satzes.
Ber. Verh. Schs. Akad. Wiss. Ldpzig 80(1928), 503507.
Hamilton, R. S. [I] Extremal quasiconformal mappings with prescribed boundary

values. Trans. Amer. Math. Soc. 138 (1969), 399406.
Keen, L. [1] On Fricke moduli. Advances in the Theory of Riemann Surfaces. Annals
of Math. Studies 66. Princeton Univ. Press (1971), 205224.
Gesammelte mathernatische Abhandlungen, Vol. 111. Julius Springer
Klein, F.
(1923).
Kobayashi, S. (1] Hyperbolic Mantfolds and Holomorphic Mappings. Marcel Dekker

(1970).
Kra, 1. [I] On Teichmller spaces of finitely generated Fuchsian groups. Amer. J.

Math. 91 (1969), 6774.
[2] Canonical mappmgs between T'eichmller spaces. Bull. Amer. Math. Soc.
N. S. 4(1981), 143179.
Kraus, W. (ii Ober den Zusammenhang einiger Charakteristiken elnes einfach zusammenhngenden Bereiches mit der Kreisabbildung. Mitt. Math. Sem. Giessen
21(1932), 128.
Kravetz, S. [Ii On the geometry of Teichmller spaces and the structure of their
modular groups. Ann. Acad. Sci. Fenn. A I Mash. 278 (1959), 135.
Kruschkal, S. L. and Khnau, R. [1] Quasikonforme Abbildungenneue Methoden
und Anwendungen. Teubner-Texte zur Math. 54, Leipzig (1983).
S. L. (I] Teichmller's theorem on extremal quasi-conformal mappings.
Siberian Mqth. .L. 8(1967), 231244. Translated from Sibirsk. Mat. 2. 8 (1967),
313332.
KUhnau, R. [1] Wertannahmeprobleme bel quasikonformen Abbildungen mit ortsabhngiger Dilatationsbeschrnkung. Math. Nachr. 40(1969), 1il.
[21 Verzerrungsstze und Koeffizientenbedingungen vom Grunskyschen Typ
fr quasikonforme Abbildungen. Math. Nochr: 48 (1971), 77105.
Kuusalo, T. [I] Boundary mappings of geometric isomorphisms of Fuchsian groups.
Ann. Acad. Sci. Fenn. A I Math. 545(1973), 17.
Math. Soc.
Lehner, 1. (1] DiscontinuoUs Groups and Automorphic Functions.
(1964).
[21 Automorphic forms. Discrete Groups and Automorphic Functions, ed. W. J.

Harvey. Academic Press (1977), 73120.
Lehtinen, M. [1] A real-analytic quasiconformal extension of a quasisymmetric function. Ann. Acad. Sci. Fenn. A I Math. 3(1977), 207213.
[2] On the inner radius of univalency for non-circular domains. Ann. Acad.
Sd. Fenn. A I Math. 5(1980), 4547.
(31 Remarks Qfl the maximal dilatation of the BeurlingAhifors extension.
Ann. Acad. Sci. Fenn. A IMath. 9(1984), 133139.
[4] Estimates of the inner radius of univalency of domains bounded by conic
sections. Ann. Acad. Sci. Fenn. A I Math. 10(1985), 349353.
inner radius of univalency. To appear in Ann. Acad. Sci.
[5] Angles
Fenn. A IMath. it:Lehto, 0. (1] Schlicht functions with a quasiconformal extension. Ann. Acad. Sci.
Fenn. A IMath. 500(1971), 110.
(21 Group isomorphisms induced by quasiconformal mappings. Contributions
Bibliography
251
to Analysis. A collection of papers dedicated to Lipman Bets. Academic Press

(1974), 241244.
131 Quasiconformal mappings and singular integrals. Jstituto Nazionale di

Alta Matematica. Symposia Maiheinalica XVIII. Academic Press (1976), 429
453.
[4J Quasiconformal homeomorphisms and Beltrami equations. Discrete Groups

and Automorphic FunctiOns, ed. W. J. Harvey. Academic Press (1977), 121142.
[5] On univalent functions with quasiconformal extensions over the boundary. Journal dAnalyse Math. 30(1976), 349354.
[6) Domain constants associated with $chwarzian derivative. Comment. Math.
HeIv 52 (1977), 603610.
[7] Remarks on Nehari's theorem abo& the Schwarzian derivative and schlicht
functions. Journal d'Analyse Math. 36(1979), 184190.
Lehto, 0. and Tammi, 0. [1] Schwarzian derivative in domains of bounded boundary rotation. Ann. Acad. Sd. Fenn. A I Math. 4(1978/79), 253257.
Lehto, 0. and Virtanen, K. 1. (1) Quasiconformal Mappings in the Plane. SpringerVerlag (1973).
Martio, 0. and Sarvas, J. [1) Injectivity theorems in plane and space. Ann.Acad. Sci.
Fenn. A IMath.4(1978f79),383401.
Maskit, B. [1] On boundaries of TeichmUiler spaces
on kleinian groups: II.
Annals of Math. (2) 91 (1970), 607639.
Mon. A. [I] On quasi-conformality and pseudo-analyticity. Trans. Amer. Math. Soc.

84 (1957), 5677.
Money, C. B. [1) On the solutions of quasi-linear elliptic partial differential equations. Trans. Amer. Math. Soc. 43(1938), 126166.
Mostow, 0. D. [11 Strong rigidity of localiy symmetric spaces. Annals of Math.

Studies 78. PrincetonUniv. Press (1973).
Narasimhan, R. [1] Several Complex Variables. Univ. of Chicago Press (1971).
Nehari, Z. [1] The Schwarzian derivative and schlicht functions. Bull. Amer. Math.
Soc. 55(1949), 545551.
[2] Conformal Mapping. McGraw-Hill (1w).

Nevanlinna, R. Ill
Springer-Verlag (1953).
Newman, M. H. A. 111 Elements of the Topology of I4fane Sets of Points. Cambridge
Univ. Press (1961).
Paatero, V. [1) Uber die konforme Abbildung voif Gebieten deren RAnder von
beschrnkter Drehung sind. Ann. Acad. Sci. Fenn. Ser. A XXXIII: 9 (i931),
1 79.
Pfluger, A. [1] Quasikonforme Abbildungen und logarithmische KapazitIt. Ann. Inst.

Fourier (Grenoble) 2 (1951), 6980.
[2] Uber die Konstruktion Riemannscher Flchen durch Verheftung. /. Indian Math. Soc. 24(1961), 401412.
Pommerenke, Chr. 111 Univalent Functions. Vandenhoeck & Ruprecht (1975).
Reich, E. [1] On criteria for unique extremality of Teichmller mappings. Ann. Acad.
Sci. Fenn. A lMath. 6(1981), 289301.
Reich, B. and Strebel, K. [1] On the extremality of certain Teichmller mappings.
Comment. Math. Helu. 45(1970), 353362.
[2] Extremal quasiconformal mappings wfth given boundary values. Contributions to Analysis. A collection of papers dedicated to Lipman Beis. Academic Press (1974), 375391.
Riemann, B. [1] Grundlagen fr eine ailgemeine Theorie der Functionen einer vernderlichen complexen Grsse. Inauguraldisscrtation, Cittingen 1851. Gesammelte mathematische Werke und wissenschafthicher Nachlass, 2nd ed. Teubner,
Leipzig (1892), 348.
Royden, H. L. [1] Automorphisms and isometrics of Teichmller space. Advances in
Bibliography
iheory of Rieniann Surfaces. .4nnais of Math. .S'zudies66. Princeton Univ.

(1Q71).
383.
121 Intrinsic metrics on Tcichmller space. Proc. International

Vancc'uver 1974, vol.2. Canadian Math. Congress (1975), 217
Schitfer, M. jlJ A variational method for univalent quasiconformal mappings. Dukc

J. 33(1966), 395411.
M. and Schober, G. [1 Coefficient problems and generalized Grunsky

functions with quasiconfonnal extensions. Arch. Ra:ionii
for
Anal. 60 (1976), 205228.
11] Univalent
Spnngcr -Verlag (1975).
A. (!)
).
Rn;
Selcctcd Torics. Lecture Notes in Math.
cinige Abhildungsaufgabcn. J. fr reine und angewandte
70(1869), 105120, or Gesonimehe Mathemotische ilbhandiungen. Vol. II,

Springer, Berlin (1890), 6583.
Characterization of quasi-disks and Teichmller spaces. Thoku Math.
.'7 (198.5). 541- 552.
Ci (lJ Introduction to Riemann
i\ddison-Wesley (1957).
M:(1) Sinaular Inregrals arzdDilTerenticbiliiv Proprrues of Functions.
Press (1970).
Zur Frage der Einder'tigkeit extremaler quasikonformer Abbildungen

des Uinheitskreisec. Comment. Mash. llelv. 36(1962), 306323.
quadratic differentials.
On the
structure
of Math. Studies 79. Princeton Univ.

(irc.ups and Riemarin Surfaces.
(1974). 419438.
of exircinal Techmudllcr mappings. Journal dAnairse
On the
.tfath. 30(1976). 464-481)
[41 On lifts olextremal quasiconformal mappings. Journaid Anakse Math. 31
(1977), 191203.
(5] On quasiconlormal mappings of open Riemann sur ices. Comment. Math.

lIe/i'. 53(1978), 301 -321.
[61 Quadratic f)jffereneials. Springer-Verlag (1984).
reichmuller, 0. (1] Extremale qua sikonforme Abbildungen und quadratische Differentiale. .4bh. Preuss. Akad. Wiss., math.-na:urw. K!. 22 (1939), 1197, or
Ge,cammelge AbhandlungenCoI!ecred Papers, ed. by 1. V. Ahifors and F. W.
Gehring, Springer- Verlag (1982), 335531.
L21 Beslimmung der extremalen quasikonformen Abbildungen bei geschlossenen orientierten Riemannschen Flchen. Abh. Preuss. Akad. Wiss., math.-
nuurw. KI. 4 (1943), 1 --42, or Gesammelte AbhandlungenCol!ected Papers, ed.

h, L. V. Ahlfors and F. W. Gehring, Springer-Verlag (1982), 635676.
Vernderliche Riemannsche Flchen. Deutsche Math. 7 (1944), 344359,
or Ge.vammelle Abhandlunyen
Papers, ed. by L. V. Ahifors and F. W.
Gehring, Springer-Verlag (1982), 712727.
Thurston. w. p. 111 Zippers and schlicht functions. To appear.
Tukia, P. [I] Quasiconlbrmal extension of quasisymmetnc mappings compatible with
a Mhius group ..4ta Math. 154 (1985), 153193.
Vekua. 1. N. (I) Generalized Analytic Functions. Pergamon Press, Oxford (1962).
Weyl, Ii. [I] Die Idee der Riemannschen Flche..Teubner, Leipzig u. Berlin (1913).
2uravkv. 1. V. [I] Univalent functions and Teichmller spaces. Soviet Math. DokI. 21
(1980).
Translated from Doki. Akad. Nauk SSSR 250 (1980), 1047
1050.
Index
Abikoff, W. 2, 192, 198, 204, 221, 247

Absolutely continuous on lines 20
Ahifors, L.V. I, 3,
12, 16, 26, 33,
34, 35,
36,
40, 49, 72, 8), 88, 116,
173; 182, 207, 226, 232, 247
AhlforsSario
3.
128, 135, 136, 137,
140, 141, 142, 143, 147

Arc condition 43
Area theorem 58, 78
Caldern 26
inequality 26
Calvis, D. 126
Canonical mapping of a quadrilateral 7
Carleman, T. 226
CaldernZygn1und
Class
338,
S 59
79
Astala,K. 117
73
Beltrami differential 133

Beltrami equation 24
existence theorem 28, 177
theorem 24, 177
Bers, L. 1, 2, 23, 24, 72, 88, 89, 98,
99, 111, 116, 144, 182, 191, 199,
203, 204, 207, 211, 226, 232, 239
Bers unbedding 114, 203
Beurling, A. 8, 16, 33, 34, 36

BeurhngAhlfors extension 33
Bieberbach 58
Biholomorphic 206
Bojarski 27
Bordered (Riemann) surface 131, 132
Boundary rotation 62
Bourbaki, N. 206
Complex. analytic structure 130

Complex dilatation 23, 177
extremal 105
trivial 19)
Conformal structure 130

lifted 142
projected 142
Contractible space 109
Cover transfonnation 138
Covering group 138
transitive 138
Covering surface 137
branched 137
normal 136
smooth 135
universal 137
unlimited 135
Index
Deformation equivalent 185

Differential 132
Abelian 132
Beltrami 133
foragroup 148
infinitesimally trivial 224
(m,n)- 132
quadratic 132
trivial
191
quotient 19
Dirichiet region 151
Distance
between domains 67
between quasisymmetric functions 108
from a disc 61
hyperbolic 5
Teichmller 103, 104, 183, 184
Dilatation
105
8 61,67
q 103
p 108
103, 104, 183, 184
;190
Distortion function X 15
Divisor 159
Douady. A. 194
Double of a Riemann surface 155
DunfordSchwartz 26, 222, 229
Duren, P.L. 77
Gauss, C.F.
24, 128, 135, 144
Gehring, F.W. 21, 23, 38, 41, 49, 84,

90, 92, 94, 105, 116, 117
Generalized derivatives 21
Genus 158
Geodesic
of quadratic differential 170
of Teichmller metric 105, 184
Golusin 79
Gronwall 58
Grothendieck 2
('irotzsch,H. 11,20
ring domain 11
Group (see also Mbius group)
discrete 153
elementary 152
elliptic modular 213
Fuchsian 150
fundamental 136
Kleinian 153
limitpointof 140
limit set of 151
modular 188
of first (second) kind 154
properly discontinuous 140
quasi-Fuchsian 155
of right translations 99
set of discontinuity 152
universal modular 99
Orunsky 79
Earle, C.J. 2, 34, 109, 194, 199, 226

EeIIs, J., Jr. 34, 109
Elementary group 152
Equivalent
Beltrami differentials 184
complex dilatations 97
quasiconformal mappings 97, 182
Fhnn,B. 117
Ford, L.R. 152
Fricke, R. 156, 241
Fuchsian group 150
Fundamental domain 149
Fundamental group 136
Gardiner, F.P. 102
Hamilton, R.S.
sequence
228, 232
229
Hersch 13
Hubert transfonnation 25
Hue 90
Holomorphic function 51
in a Banach space 206
on a Riemann surface 131
Hornotopy
lifting property 147

of mappings 146
modulo the boundary 180
Hyperbolic
distance 5
metric 5, 6, 148, 246
sup norm 54
255
index
Induced group homomorphism 145

Inner radius of univalence 118
Isothermal coordinates 134
differentiable 129
real 129
Maitio, 0.
49, 52
Maskit,B. 198
Jordan domain 7
Keen, L.
241
Klein, F. 135, 156

Kleinian group 153
Klein's method 156
Kobayashi, S. 246
Koebe 144
function 59
one-quarter theorem 6
K-quasiconformal 12, 176
Kra, I. 2, 194, 226, 229
Kraus, W. 60
Kravetz, S. 221
Krushkal, S.L. 79, 2Z. 228
KUhnau, R. 72, 74, 75, 78, 79
Kuusalo, 1. 195
Lebesgue measurable mapping 130
Lehner, J. 3, 129, 149, 152, 153, 154,
156, 213 216, 224
Lehtinen, M. 34, 36, 37, 122, 124, 125,
126, 127
Lehto, 0.
17, 21, 23, 28, 65, 66, 71,
72, 74, 76, 78, 79, 122, 199
LehtoVirtanen 2, 4, 8, 9, 10, 11, 12,

13, 15, 16, 18, 20, 21, 22, 23, 25,
26, 27, 28, 29, 33,
38,
48, 101
Lift
of a differential 148
of a mapping 145
'. ofa path 135
Limit point of a group 140
Limitsetofagroup 151
Maximal dilatation
of a quasiconformal mapping 12
of a quasisymmetric function 31
Meromorphic function 51, 131
Mirror image of a Riemann surface 155
Group)
Mbius group
Dirichiet region 151
invariant domain 155
normal polygon 158
quasiconformal deformation 191
Modular group
elliptic 213
of a Teichmller space 188
universal 99
Module
ofapathfamily
12
a quadrilateral 8, 9
of a ring domain 10
Monodromy theorem 136
Mori, A. 21
of
Morrey,C.B. 24
Mostow, G.D. 195
Narasimhan,R. 211
Natural parameter 162
Nehari, Z. 6, 60, 90, 124
Nevanlinna, R. 143, 144, 158
Newman, M.H.A. 7, 15,44,85
Norm
hyperbolic sup- 54
of Schwarzian derivative 54
Normal family 13
Normal polygon 158
Linearly locally connected 44

liderivatives 21
Ohtsuka 143
Osgood, B.G. 49
Outer radius of univalence 66
Majorant principle 77
Paatero,V. 63
Manifold
129
Parameter
206
Pifuger, A. 12, 21, 23, 49, 101
Banach
complex analytic 130, 207
Poincar
transformation 129
144
Index
256
Schiffer, M. 79
density 6,148
nietric6
series 223
Pommerenke, Ch. 79, 90, 92, 94
theta
Quadratic differential 132

critical point 161
initial 235
metric induced by 168
natural parameter 162
orientable 164
regular point 161
-.
Schobcr,G. 79
Schwartz, J.T. (see Dunford-Scbwarrz)

Schwarz, H.A. 52, 137
derivative 51
Schwarzian domain 83
Sewing problem 100
Shiga,H. 204
Sobolev space 21
Space
A(G) 222
A (1) 223
terminal 235
I? (G). L' (G) 222

N(G) 224
trajectory 163
Quadnintcrall
conjugate 46
Quasicircle 38
Quasiconformal deformation 191
Quasiconformal mapping 12, 176
analytic definition 22
diffeomorphism 19
extremal 37, 183
geometric definition 12
I-dimensional 31
of a Riemann swface 176
Quasiconforinal reflection 39
Quasidisc 38
Quasi-Fuchsian group 155
Quasisyrnmetric functions 31
space of 108
Quasisymmetry constant 32
Reduced Teicbrnller space 183

Reflection principle 46
Reich, E. 37, 227, 229, 230

Rengel's inequality 9
Riemaun 130, 135
mapping theorem 143
space 182
surface 130
Riemann-Roch theorem 160
Ring domain 10
Royden, H.L. 2, 226, 246
Sario, L. (see AhlforsSario)
Sarvas, J. 49, 52
Ill
Q(G)'197
Q(1) 197
R, 182
T 97
T(G) 198
T(1) 114
7', 212
7',
182
U 115
U(G) 197
U(l) 198
X 97, 108
X(G) 194
Springer, G. 3, 129, 151, 158, 159, 160
Stein, E.M. 26
Swflov 138
Straight line 163, 242
129, 166, 167,
Strebel,K.
169, 190, 227, 229, 230, 232
Study 63
Surface 130
of finite type 247
Tamnu,O. 65
Teichmller 1, Il, 103, 163, 170, 174,
181, 182, 207, 216, 226, 232, 233,
239
disc 245
distance 103, 104, 183, 184
existence theorem 232
imbedding 241
lemma 170
mapping 231
index
nng domain 11
space 97, 182
uniqueness theorem 237
Thurston, W.P. 2, 118, 204
Universal covering surface 137

Universal Teichmller space 97
Tienari 49
Visl, 3. 13, 17, 38
Tonis 150
Total set 206
Vekua, 1.N. 26, 28
Trajectory of a quadratic differential

critical 164
cmss-cut 167
horizontal 163
periodic 166
spIral 161
Tulcia, P. 110, 194
Virtanen, K.1. 17 (see also Lehto

Virtanen)
Weili, 0. 88
Weyl, H. 130, 135
YOjob 23
Uniform domain 41
Uniformly equivalent 201
Univalent function 52
LV. 204
Zygmund 26

109 - UnivalentfunctionsandTeichmuellerSpaces PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

109 - UnivalentfunctionsandTeichmuellerSpaces PDF

Uploaded by

Copyright:

Available Formats

Olli Lehto

AMS Subject Classirications: 30-01, 32615, 30F30. 30F35

Library of Congress Cataloging in Publication Data

Univalent functions and Teichynller spaces.

1987 by Springer-Verlag New York

Reprinted by World Publishing Corporation, Beijing, 1990

Quasiconformal mappings and the Beltrami equatior

4.3 Singular integrals

4.4 Representation of quasiconformal mappings

(Jnivalence and Quasiconforrnal Extensibility of Meromorphic

Universal Teichmiiller Space

5.4 Inner radius and quasiconformal reflections

1. Manifolds and Their Structures

6.4 Trajeciodes in the large

6.5 Periodic trajectories

4.7 Ben unbedding

5.2 Banach manifolds

theory of Teicbmller spaces studies the different conformal structures

modulo conformal mappings.

of quasiconfonnalmppings iii the plane.

The development of the theory of Teichmller spaces along these lines

with provide an introduction to the Teichmller theory, they leave aside

knowledge is needed from the theory of quasiconformal mappings and of

Chapter I is devoted to an exposition of quasiconformal. mappings. We

Chapter 11 deals with problems of univalent functions which have their

The philosophy of Chapter IV on Riemann surfaces is much the same as

characterizations and, guided by the results of Chapter III, develop their

classical complex analysis. We expect the reader to be familiar with the

faces would be helpful, but is not meant to be a necessary condition for

self-mapping of the halfself-mapping of the real axis to a

'1. ConfOrmal Invar ants

In the firstthree chaptets of this monograph, we shall be

with mappings whose domain and rang are subsets of the

In addition to the euclidean and spherical metrics, we shalt repeatedly

which map D onto itseli By

undet the group of conforthal mapings of

defines a'metric which is

The Riemann mapping theorem says that every simply connected

Then the differential

For domains A not containing oo we also have the lower bound

This is proved by means of the Koebe one-quarter theorem (Nehari [2],

maps A onto a domain A' which does not contain 0 or

conformal invariance of the hyperbolic metric and by (1.4),

Since d(f(z),8A') If(z)I, the inequality (1.5) follows.

in V.9.6 we define the hyperbolic metric on an arbitrary complex analytic

1.2. Module of a Quadrilateral

can, modulo orientation, be prescribed arbitrarily on the boundary of the

Let f be a conformal mapping of Q onto a eucidean rectangle R. If the

boundary correspondence is such that f maps the four distinguished points

monograph, ro which several references will be made in Chapter 1, we thall

conformally invariant, i.e., if f is a conformal mapping of a domain A and

1.3. Length-Area Method.

In order to arrive at this characterization of the module we consider the

We can get rid of the canonical mapping f if we introduce the family P

given p e P a function Pt in the canonical rectangle R by

du(J' Pi(U + iv)dv).

1.4. Rengel's Inequality

the quadrilateral. For a precise formulation of the result let us consider a

Q for every n and

to every r > 0 there corresponds an ; such that for n

zfl)) = M(Q(z1, 22, Z3, Z4)).

To prove this, we only need to carry out a canonical mapping of Q(z1,z2,

Rengel's inequality also makes it possible to characterize conformality in